Feb 21, 2026 Enterprise Brain: UI Workshop - Transcript 00:00:00 Satyasri Prabhakar Mantripragada: continuous half hour. No. Rajashekar G: Listen. Satyasri Prabhakar Mantripragada: Okay, let's get started. Okay, I like giant. Okay. getting started. Yeah. Uh you can see now those who are Gopal Gottumukkala: Yeah, we Satyasri Prabhakar Mantripragada: online. Okay. So uh now this is a five hour long workshop uh or strategy discussion whatever we call it as uh I have I wanted to quickly go go through all the agenda how we wanted to approach this so this is the so this is the timeline so first I'll give you a brief on different use cases so that you can consume the first half an hour on what we want to achieve uh all the use cases at the high level and then each of the departments will plan for the till up till 2:00 on how do they accomplish the use cases and if you need any further clarifications on the use cases you can ask in the first half an hour itself you can always pitch in whenever you need and then between 12:30 to 2 you'll finalize on the approach uh it's up to your team on how you wanted to finalize that but the use cases need to be accomplished. 00:02:03 Satyasri Prabhakar Mantripragada: We'll break at 2:30 2 to 2:30 and post that the finalized approaches will be brought to the table by each of the departments with an objective to accomplish accurate information presentation reliable system which is secure and scalable. This is what we need to accomplish. Once we finalized all this by all the departments then we define the what are all the non negotiable con constraints like performance data integrity etc. If there are any we add so once these are done the final plan is get back to work on action items for each of the teams along with the timelines. Okay. Next. Hello. So, first is uh I wanted to give you an overview on so far we have only seen Gmail integration, Salesforce integration for all of us to get to the same uh table. I wanted to give you an overview of the the the bigger picture of uh enterprise brain uh or operation meta is that we wanted to develop a system which facilitates information consolidation from across different sources like these like emails cloud stoages meetings scheduulers application databases internet chat spaces meetings is duplicated ERP or any other enterprise systems I'm not going to specifically call it out whether it is project management ment systems or HCM systems or CRM everything is categorized under 00:03:43 Satyasri Prabhakar Mantripragada: reacts and then Jira audio files, video files, images, documents and there can be so many other forms which can be input to it. It captures that information, processes it and then gives you information for based on your query or based on its intelligence. And the application is so adaptable so that you can just plug and play with the new application or new system with minimal effort. And uh it should also offer personalization features so that it can on demand learn on the fly as it as the application is being used by the users or it can learn from an existing system or existing database. Uh the primary factor for this is it should be speed and accurate uh with respect to giving responses to the user queries and it is very secure and trustable relig Okay. Okay. Okay. Maybe I I'll get that screen. Rajashekar G: Shelton. Satyasri Prabhakar Mantripragada: Okay. Fine. One minor note. So give us one minute mu I can share you you present this. 00:05:22 Satyasri Prabhakar Mantripragada: Okay sir things first. I'm I'm connected. I'm just saying. Okay, I'm joining. You got it. Naveen Puttagunta: books on my little picture. Okay, just to you know just wanted to chime in with the the first bullet point is information consolidation. The goal is actually not to consolidate information. We are not actually going to consolidate the information at all. Information sources will be wherever they are. We will pull that information to be able to build uh our knowledge graph for this platform but not to actually oh let's pull all that information and put it in one place. That is not the goal at all. That is that kind of stuff is like data lake and other things. This is this the goal of this system is not to make sure that all the information is in one place. The main goal of this information is to be able to build a knowledge base of the entire information available everywhere. Two different things. 00:07:39 Naveen Puttagunta: Got it. Because from a data perspective, understand that we are not putting and creating a one more data lake or here. But we are understanding by the statement. We some information is coming from an email. Some information is coming from a meeting. Yes. But both are talking about a similar project. We are able to consolidate and tell a common story. Consolidate is a wrong word but because we we understand it so that we are able to uh uh you know build our knowledge about those different things together knowledge is another thing wisdom is what I would say it's a wisdom but ultimately we telling that there is some information happening online now so consolidation it's a flat information has no meaning Right. Wisdom and you you you understand the information conclude and do something about it and there is a connectivity that you are establishing information. Yesd they are interconnected and there is some inference we are calling out of that connection. Okay that is what he's trying to say doesn't want to call it consolidation. 00:08:55 Naveen Puttagunta: Okay, this you have something or I have something one minor clarification right just I think this this aspect is important to understand suppose you have a meeting transcript right so by processing it and by storing embeddings of that meeting transcript the AI's knowledge graph understands something about that meeting. Not that it can it cannot it may not be able to reproduce it word for word. If that re word for word quotation reproduction is needed, you still have to go back to that source and pull that as a reference and show you right that is but we understand the the knowledge graph understands that oh this meeting was about uh the vision for AI uh the meeting was about uh operation Mha. So it will understand that it may even have the linkages necessary to say okay this was what was discussed. Yes. But not like a reproduction is not for that you have to go back to the question. Yeah. Okay. Venkatesh Tammareddy: Uh then I have one uh uh question or one I need I mean uh I I understood little bit differently actually knowledge graph is different and embeddings are different. 00:10:30 Naveen Puttagunta: Okay. Venkatesh Tammareddy: So knowledge graph is will talk about the relationship between the entities that we have in the system. Whereas embeddings basically what they talk about is like for example in the what whatever transcript that you have described. Naveen Puttagunta: Hm. Venkatesh Tammareddy: Yes. Uh that will be stored as an embedding. Like for example uh one example that I can give you is like for example there is a meeting that happened on a particular feature and uh that particular feature has been divided into stories and epics whatever it is. Now uh the fe the features and epics are related uh will be represented in the knowledge graph saying that you know these particular features are associated or these stories are associated to this feature and this feature belongs to this epic that is the information that the knowledge graph holds. Whereas on the embedding side uh whatever the transcription that we are going to I mean the transcription we will convert it into an embeddings. So that like for example when I ask about give me an information about this particular feature then what it does is I mean basically how it is how it will happen is now it will talk to the knowledge graph and understand the relationship as well as uh it will 00:11:47 Naveen Puttagunta: Stop. Venkatesh Tammareddy: also look at the embeddings that we have reading whatever the user is trying to ask. It combines those two and then uh it will give I mean it will send for inference and it will it will give it will give us back the uh uh data or whatever uh the data format can be that is what what I understood at least embedding this knowledge graph. Naveen Puttagunta: fair enough. Um yeah Manisha Rajar do you have enough uh insight into that to confirm or change that? No sir award itself. Yes. But uh does that mean that when I process that transcript you will also add whatever is coming out of that uh transcript and meeting to the knowledge graph automatically and also store you store as embeddings. Or are they two different concepts? Venkatesh Tammareddy: uh they they are two different concepts actually like for example Naveen Puttagunta: No. Venkatesh Tammareddy: uh if if I take another example uh let's say uh um there are two different concepts Actually knowledge graph talks about the representation I mean the interconnection between how those elements are connected. 00:13:11 Venkatesh Tammareddy: On the other hand embedding stores the meaning of what I am actually trying to describe. It can be a textual format, it can be an image or it can be any form basically. So it stores the meanings of it. Like now if I if I ask about give me the performance of banker it it will search the complete uh details. How is banker connected to the system? uh what is he working on that will come from knowledge base and if there is any feedback that is already given on ward by someone uh right that will also that will look into the embeddings and then a prompt will be formed and uh basically that is how the output will be so they are two different uh things uh when we are talking about knowledge graph and embeddings Naveen Puttagunta: understood. So but my my question is more of so when we process an email uh do we u store the not only do we store the embeddings but we also kind of update the knowledge graph. Is that happening or is that what will happen or we'll only do one and not the other? 00:14:18 Venkatesh Tammareddy: uh I I mean at least that I think on uh I think it varies to use case which is use case. Now uh in knowledge graph like in the if you take the same example of an email Naveen Puttagunta: Okay, we've lost you, but can you answer that Manisha sir? Yeah, sir. Okay sir. So basically embedding contain sorry basically what is embedding is if there is a text like information or email the complete no I understand embedding versus I'm asking in which cases do you update only one and not both or do you always update both? I mean we always update both sir because every text should be converted into an embedding and then it will be processed into the graph into knowledge graph. So you you you do both. Okay. So embeddings and knowledge graph are both updated. But even before that my my two cents this feels this whatever you guys are talking about highlighting email meetings and this I have not I have not been able to put it properly but it feels like we are building something for the that is for the enterprise ecosystem that we wanted to connect with the data sources can be anything I I um I I mean the examples that are being discussed the the couple of example because somebody you know is just talked about email we jumped into email but um think of connecting to ERPs think of connecting to Oracle unifier right you know the the actually the the best 00:16:17 Naveen Puttagunta: use case is probably you know if you look at Tata Steel's the brief that they gave it is fantastic that is what the the um core of enterprise brain what we're trying to deliver perfect example of that now can I talk sure it still feels like that the reason is since day one whenever we have been talking about enterprise brain every time meetings email chats come along with ERP and all of it okay okay first of all people you know there is a privacy concern there when it comes to chats, emails and all of it. So the grouping this is what I was talking about yesterday that first we need to identify the grouping. One way of grouping that I was talking about yesterday is structured unstructured streamed and all of it. But you also have to come up with layers where there is business data, personal data, when, what, and all of it. Like when you put it up like this, right, people I mean uh this is not a sales presentation. One second. I'm not criticizing any I'm just saying that this is a from a PO standpoint. 00:17:38 Naveen Puttagunta: That's PO standpoint. I want the priorities to be changed is my request. See we are focusing too much on email meeting I mean at least Jira though I don't know how much see Jira is for developers we are saying one level CXO and then we are talking about Jira right so for me I I feel like our thinking in terms of something needs to change for me first what are the I mean I'm not able to put it forth and I want you guys to understand my concern where I'm coming from rather than challenge that it is okay because I have a concern and I'm not able to express it properly. We need to first understand all the B2B um elements B2B people will be willing to freely give permissions to it like Tatastel for example it'll give permission to Premier it'll give permission to Salesforce it'll give permission to SAP and all of this there they won't have any issues what is that layer then there are other kinds of layers like video I don't know streaming of camera information I I mean I'm just talking out loud those are the next one last is emails and meeting access and chats and probably Jira will come somewhere in between Jira is I mean I don't know how else to put it so this is my input fair enough well take it and for me when you guys are working the reason I'm saying that is all of like I mean when when uh you guys came back and said I'm working on integrators for email integrators for chat I'm not 00:19:32 Naveen Puttagunta: interested in integrator for email and chat right now because I it takes a huge amount for anybody to be doing that they're not nobody will be willing to do it like that but if you say if I build an integrator for SAP by now as an example I can easily go to the market not mean that you guys have to figure that out. So first of all for me I don't know who is going to own this task this this I need I need one of you to break down the different types of data into multiple I mean look at it from a different perspective who praer and I need data types from different angles that's the first task we need to work Um um your point is well taken Pritma and uh this is not a justification. I'm saying we fell into the uh trap of hey let us search for keys where there is light. I know because we have access to it, we're doing it. We have access to Jira, we we are doing the Jira agent access to Salesforce Salesforce agent because we have access we're doing it. 00:21:05 Naveen Puttagunta: So we have access to that was the first one that we did. In fact, if you observe you can you do chat GP in all the B2B spaces what is most used the Microsoft suit or Google suite it shows that it is Microsoft suite but I'm not again I'm not debating what I'm trying to say is if we are getting into enterprise places looks like Google again I could be wrong this is my analysis I definitely could be wrong. Google is used in small and midsize companies. All most of the enterprises use Microsoft suite. So you said everything is Google based. complete. So somewhere I feel like I need to revisit all of this again. Pratima, this has been thought through. It's not that we just took a trivial uh decision to access do it on Google because when you are talking about enterprise brain and you want a realistic demonstration of the capability you know you shouldn't I mean there is simulation you can always simulate data and but then people will say that is simulated data where is the real data you you know our real data that you can show outcomes of hey this is what we can enable. 00:22:49 Naveen Puttagunta: This is what you can accomplish can only be done on real data realistic data and you just think about where is realistic data available to you that is where the answer will be applied to now you would always do it slightly differently now and if you need Microsoft suite then we need to go figure out a partner with whom we can do now I'm not debating it so far I'm okay because what I understood was it was trial and error now that we saying let's get serious about enterprise brain. I'm requesting all of us to revisit how this this has to be done. Um if you have a different opinion we can do that PMA but please what I'm saying is understand why we did something and you can't just say oh why did you do it that way I'm saying you know you are saying you took the wrong call we are saying it has been thought through and we took the call said we took the wrong call I'm saying can we revisit there's a difference sure understood Okay, moving on. 00:23:58 Naveen Puttagunta: Uh, any other points on this? Anybody online have anything to say on this? Rajashekar G: No sir, I'm good. Naveen Puttagunta: Okay. Gopal Gottumukkala: No, no, I'm I'm good. Naveen Puttagunta: I don't think join. Okay. So first thing is we need to identify the key business critical parameters that we wanted to present. You can call it as a dashboard or uh or the landing page whatever. So these are some of the metrics at the high level that we wanted to have from a pure standpoint. Something that is more growth focused for the se uh executive. Something that needs immediate attention. a comparison across a certain department or group or region. Uh and then certain things that they need to make quick decisions, quick actions like approvals uh or quick actions on certain things. Maybe they need to schedule something and something is very dynamic which monitors the external world or something and then it fetches that information for to be consumed by the CE executive. Um I have a question here and um and not a question but one of the things that that has been running in my mind is uh we are building it as a platform uh and when we are selling this when we are going to sell it for multiple types of domains uh h how are we handling that part so first of all I just want to understand what is the what are the domains that we targeting because different domains has different types of uh these things. 00:25:52 Naveen Puttagunta: How are we going to generalize them? To me, I would suggest it should be domain agnostic and then say the platform is horizontal. The platform has certain capabilities like okay now we are not a product company, right? We are a services company. So I am going to look at where the opportunities are where did they come from for example the first opportunity came from consent saying that we should have a Salesforce agent so we did actually that that it started as NLSQL right but NL2SQL morphed into enterprise brain so there was an evolution there in thought process saying that actually there is very only a small sliver of value turning natural language into queries and then just getting results. That morphed into no no the the bigger vision is enterprise brain where you are able to uh support better decision intelligence by adding more capabilities. So at the platform level enterprise brain has to bring some capabilities and some uh u some some capabilities and some benefits and outcomes then okay then I'm looking at the the our sales engine to say where are our opportunities the first opportunity IRM so we do an IRM agent second opportunity is was salesforce so we do a salesforce agent now the third opportunity is with Tata steel with several sources. 00:27:32 Naveen Puttagunta: So we're going to do those sources, right? So it is always going to be customized for whatever the target use cases. Is there should we have a very specific focus area and go after it. I will leave that to the sales team to say yes, we let's focus on this or focus on that and then we can act accordingly. That is what I would say. Uh I have few questions man. Sure. Uh now I mean the reason I'm asking it is because I need that clarity. So uh so we say that you know we we targeted salesforce agent and we targeted whatever is required for they're specific to that particular thing. Now even before we go into that step what I'm trying to understand is what is this platform going to do? Whatever we are building, what is that? Is it uh something like you know uh we are saying that you know this platform supports multiple uh kinds of data sources previous so uh yeah this is actually nice so that's I understood now my followup question on that is um like you know you take any enterprise like if you look at supply chain the way how things are organized or the domain knowledge that it ads is completely different from when you're targeting the healthcare industry. 00:28:59 Naveen Puttagunta: Now uh so what are we trying to do? How are we going to address it? Because I think that is the first step that for us to say that you know we have an enterprise frame. For me Ward I don't quite see it that way. I see that the core platform functionality is very similar across any given domain. Okay. The core platform functionality is being able to understand whatever information is out there. The very very disperate sources of information out there. Be able to connect them and they understand the interconnects between them. Be able to monitor that information uh and either proactively or reactively help the human make better decisions. right and give them better decision intelligence. Uh so now I think what I I mean how I understood is so the layer that we are trying to build is a generic layer. Yes. Uh but now if the requirement comes like for example today we have a requirement from supply chain tomorrow we might have a requirement from health sciences. 00:30:10 Naveen Puttagunta: Right. Now underneath our platform layer whatever data transformations and everything has to happen that we are going to build based on the uh customer thing. Uh yeah it's it's predom yes yes you're right you are right and we have to have so again we need to have certain theme knowledge we need to this even for concept this is what I was trying to tell them we need to understand what sales forces and salesforce documentation that's why this this thing always bothers me because no matter what we go back to email chat all or these things right that is more than Right. You have subject matter, expertise, documentation that is absolutely required. Even for optin there was this blue yonder runbook or playbook they were calling something which was index. You will have my assumption is each company will have one or many such runbooks or playbooks. So that also we need to group it into subject matter expertise. Then it has to be have multiple layers like the core industry then the client itself the company itself maybe some competitor information I don't know then the business units and to the point of individuals now to the point of individuals I could connect to zoho to get that or it could be unstructured documents of whatever form right the I at this point that is a repository that I am looking at as a group of 00:31:52 Naveen Puttagunta: documents that a company uh will give us for enterprise brain to work much smarter. So that is one layer I'm looking at wing. The next layer is the e ERP, CRM, all of these systems, HRMS systems and all of it. Then I'm looking at um um so that is what I was calling unstructured data. Unstructured data the first pool of it was what I was calling as unstructured data. Then these are all the systems they are using right uh to for their maintenance and logistics and uh execution day-to-day execution these two and the third one I'm looking at is um uh because it's supply chain and see in supply chain my my understanding is supply chain and manufacturing we may have to process a lot of either IoT information or um uh feed from cameras or images and pictures. So in definitely those two industries I see that in insurance and tech I don't see that. That's the third type of information. M the fourth information is the personal information like you know layered personal information like chat is between a group of people not it could be across the company or it could be across a few email is very very very hyperpersonalized meeting is again a group of people right so this so this is how I am looking at information for me the reason I think I've been talking about this for the last few times and I again I 00:33:52 Naveen Puttagunta: that's why even yesterday I was telling I need your help here is I don't want to go to people and say I will include your email I will include your chat that's it the next 1 hour 45 minutes is all about first of all I will not even be able to sell it right now truth be told because everybody first in fact nobody even trusts AI now we are trying to say we are building AI application that will process Ward's personal email. Do you even feel comfortable if Diwami's enterprise brain would have access to your email? You will be ex Yes. probably you. No, no, no. I'm saying corporate email is not personal. It's not personal. Please understand. But also at the same time, nobody feels comfortable with that. So you're you're losing the trust already. Okay. So basically that there will be lot of resistance. Nobody will want to get into because there are bigger problems all companies are trying to solve that email is the last thing that anybody will want to solve at this point because they don't want resistance from the rest of the employee. 00:35:04 Naveen Puttagunta: They don't want resistance from everybody at this point. Right? So for me this pool of unstructured data and these systems is what and the streaming is what I'm looking at first. This meetings email chat is a uh a different layer. Now if you yesterday we had uh two I mean we were discussing two opportunities for this AIP. One is about um uh um a gyic or I don't know if she's a gyic or I don't know if she's a therapist. She kind of gives exercise routines for pregnant women from week whatever 10 to or 8 to 40 weeks right every week she has an exercise they were asking she pretty much I told you right I was talking to you know engine or kicki people and they're asking can you build an AI engine or enterprise brain on top of this where you curate the the schedule of workout for that pregnant lady based on I'm doing fine there's nothing wrong with me. So, it's the regular one. Or I've been working out all my life and I'm pregnant now. 00:36:18 Naveen Puttagunta: So, the kind of workouts they get is a little advanced. I've never worked out but I'm okay is one type of thing. Then I have um diabetes then it's a separate thing. I have back problems or the doctor asked me to take rest. So, these have to get curated. So, they're asking can you build an AI layer and can you build an enterprise engine? Now I don't know if it is rule based. I don't know if it really requires there right when you go and say I you know I will go check your meetings check your email because it's just four five people it's all about those four five people agree but in enterprises there are much much much bigger problems that people want to solve than these okay I well taken I I'm not this is not even a debate so I you know so I I this is what I'm saying is not because I want to debate with you on this Right. Um totally email chat and this thing we will completely let us say underlay not try to make it front and center. 00:37:23 Naveen Puttagunta: Got it? Okay. No no no discussion on that. However I'm just saying please try to reconsider your thought process. I'm not saying oh we have to talk about it. I'm not saying that we'll underplay. That is it. I'm just saying in corporates email is the one thing that is absolutely monitored. It is absolutely monitor. Everybody knows that. In fact, Gmail you know change that perception for personal email. You know that your personal Gmail is being uh you know processed and monitored and that is how you get ads, Gmail ads and all of that. Right? That has changed perception. meetings. Earlier people used to be so oh meeting somebody is listening in they were so this thing now it has become oh if you don't want it you have to explicitly say by default there are at least one or two meeting agents that are listening in right it has changed the perception chat slack and things like that everybody by implicitly understands that it is being monitored why because there is an agent that the moment you type start typing something agent is processing and then doing something So some person looking at it has a different connotation than an engine processing it. 00:38:43 Naveen Puttagunta: That perception has significantly changed is what I'm saying. But I because I want you to rethink that and just consider it because we we have to be at the forefront. So I'm not but again I'm not saying oh this is the reason we have to put it front and center. I'm not going to do it. I'm going to follow your direction only. But I'm saying reconsider because it has significantly changed especially email chat drive people understand that they are being processed. It's only in small companies like ours that the email is not being monitored. In even midsize companies, email is absolutely monitored because they don't want and especially uh for people know that you know that is what they say you know please do not discuss these things on email because they are part of the record they are part of discovery process they know that they are going to get su if they get sued they need to provide all of that information they actually look for solutions to do all of these things through agents. 00:39:45 Naveen Puttagunta: So in enterprises people absolutely understand that email is actually being not personally monitored but agentic monitored. So that is just one thing I'm offering but we will completely underplay email chat and this thing you know again so we don't have to uh repeat that. So, so one action item here is when we're presenting enterprise way, let us start with like information or data layers or data source layers you're saying. Uh core is you know the uh enterprise ERP systems ERP CRM SCM systems that you are going to connect with. Then there is unstructured information in documents and uh uh things. Then there is um uh you know even custom applications all of that will come under the core databases right ERP CRM uh custom custom application databases all of that is one layer unstructured documents is another layer very media is another uh layer that may be very specific to uh industries uh and then very uh underplayed layer might be okay you have other information in your email and messages and uh meeting transcripts uh things like that maybe meeting transcripts can be put out into a different category if you're going to do this I am looking at it as four columns of information and with uh um rows at least level one level two level three uh like for example Example if it comes to database can you just write it so that we we'll just capture it. 00:41:49 Naveen Puttagunta: companies use postres some companies use lightweight then there is medium and then there is heavy like uni Oracle unifier SAP uh Salesforce these are all one one I mean they are huge now Salesforce SAP they are also going into light right so at least this is how I have been putting it in my head but But you guys can come up with something. But I want if you if we don't first look at it like this, it'll be a problem. Sorry, just to clarify what are you saying? So four columns. If you understand four columns in your mind, what are the four columns? In my mind the first one is unstructured or you know the core if you want SAP CRM. Okay. So first second column you can write them as the core business uh um the these are typically core business applications. Core business applications. Okay. Then unstructured data is all about company contextual contextual uh documents which could be subject matter expertise, it could be basled base and all of that. 00:43:26 Naveen Puttagunta: Okay. at least the following step. Uh third um um multidia content priority priority content uh videos and something these are more monitoring monitoring information. See sensor data is different from like videos audio whenever there is something associated with video pictures IoT it is a pull because it is constant streaming of information that's how in my head I'm grouping them now that's why streaming data is then the fourth one is uh um you know uh contextual uh like emails, meetings are the fourth fourth column now row level Huh? Level two. Okay. Okay. Level three business. Level two Salesforce light and they pay little and get it done. Why not? It could be an Excel sheet also. Lots of Excel sheets and stuff like that. Maybe post. Can you just write custom applications, custom custom apps, custom DBs? Matrix second second row. Um can you write uh like functional um functional packaged apps like Salesforce um like HubSpot like Zoho um you know HRMS packages things like that. 00:46:24 Naveen Puttagunta: Kinda the the third row is like full ERPs functional examples. Um um put uh see I would put Salesforce okay HRMS um Jira things like that the functional package apps I feel like a lot of research has to be done in this space for example manufacturing SAP is very very much used We need to identify in each of these things like what are we talking about and at least this is how I have been looking at information. Um then we say that these are the things that we we can correct to all of them. But what do you have? Somebody needs to identify what are the most popular ones in level three core apps again in tech. What I mean you we there is a listing everywhere and so if somebody could fill this up and as of today what do we have connectors with? We don't need connectors for everything but we need connectors for SAP. We need ready connector for Salesforce. So one thing to do you kind of show your authority that you have done your homework. 00:48:34 Naveen Puttagunta: you know exactly what it is and we can include any one of them. So naku from a UI perspective when streaming data is there the third the way my dashboard will look will be very very different from when it is not there when my email my dashboards may look very different from when it is not there so permutations and combination says I don't know 16 or 17 or I'm assuming I don't I'm not even doing the calculation I need somebody to do the calculation for How do we present? Uh did I make sense to all of you? So this HRMS or Salesforce maybe any sales cloud those are all modules of an enterprise individual modules they can be think of it this way and they are functional packaged apps so the complexity is medium when you take something like SAP the complexity is extremely high okay also has multiple modules currently and so on right and if if somebody has an ERP, right? That is why I'm saying ERP. They most likely have implemented multiple modules of that ERP. 00:50:06 Naveen Puttagunta: Okay? Manufacturing, if a if a manufacturing company implemented ERP, they're typically implementing the entire thing. Okay? You can say that each individual thing is a medium complexity one, but we're saying the highest level of complexity is in full piece at the third B. So it's basically if I understand Pratima right she's categorizing level one level two level as complexity okay level one is the easiest complex least complex level two is medium complex level three is the highest complexity is that a fair representation is that why you had level one level two level three in mind yeah so that way for example SFDC is still you know like a salesforce sales cloud marketing cloud They're all point functional areas and for me if I have to look at it enterprise light will be everything associated with level one company level I can build an enterprise light for them right and uh for enterprise companies the level three yeah so I don't I wouldn't I mean uh so how maybe we don't have to spend time but do we have to spend time on for example in the second one like this contextual documents and all of that uh is there a level one level two level three what does that mean and that's what I'm saying the way people all of us approach MAP level actually understands they may not. 00:52:11 Naveen Puttagunta: So they need some inputs um a lot more con a lot more um knowledge um about what that SAP is doing why it is what I'm so so the documentation that you will need in support of this will be high when it comes to um level three so and it may not be pure Google doc yeah actually that the column names there's absolutely no context In SAPN table names the column names key they will not it will not look like orders it will say uh SDK 3749 that will actually be the column name or that will be a table name. So you there's no relevance to it. So this gives a structure to my thinking actually. I feel like then the approach decides just it it gives us authoritative power to say we understand B2B ecosystem. Right now I'm I'm not able to say we understand B2B ecosystem anywhere. Hey, I know B2B space. I know manufacturing. I know tech space. I know this based on that. This is how I design. 00:53:42 Naveen Puttagunta: This is where I'm coming from. This is all the ecosystem you will have in terms of information any sources of information. These are the different sources of information. You can pick a subset of these and based on that our brain will sit on top of it and give you what you want. Now each one of them I want to map what type of wisdom we can give them and a wisdom neural network single wisdom connected. The more the number of uh uh wisdom points, the more complicated uh uh area gets, right? But the better their decision making frameworks will be. Yeah. Do do you guys have a Raja or Manisha? Do you guys have a shared drive or something for enterprise brain or should I do something in the IML folder? Rajashekar G: We have it Naveen Puttagunta: development. Huh? Rajashekar G: in. Naveen Puttagunta: Development and projects. Development projects. Uh, do I have access to it? Have development and QA projects. Rajashekar G: I Naveen Puttagunta: Okay. 00:54:53 Rajashekar G: will Naveen Puttagunta: Projects projects should have but one we can create on the top and all like a project. proposals. No proposals. Can you just copy paste it into this sheet? Mckenzie circle then it kind of gives us an authoritative and people will really think they know. Okay. But these guys know what they're talking. Okay. So also comes under this aspect. Yeah. I think refresh. Sorry. Are you guys seeing my screen? You're not seeing my screen. Okay. You're seeing his screen. you guys are able to see the spreadsheet. Yeah. Rajashekar G: Yes sir. Naveen Puttagunta: So this is what Pratima is categorizing as core business application, contextual documents, streaming data, um contextualized personal information. I need somebody to I need somebody to relook into this and do it. Uh I mean either either my understanding is right or you may have a better way of doing this. So, but at least I need something at least this this way. 00:57:46 Naveen Puttagunta: Okay, understood. And I need the list of all the things that fall into it so that all of us understand what is going on in the world. complement. So basically this is a scheduling tool. um um so it's a scheduling tool where drivers and then pay something I don't know if at least talk like this it makes more sense and I'm thinking Rakkesh can have access to this leadership Gopal Gottumukkala: Oh yeah. Yeah. Yeah. Okay. Naveen Puttagunta: team leadership Gopal Gottumukkala: Okay. Yeah. Naveen Puttagunta: team and then individually if one particular designer is working Gopal Gottumukkala: Yeah, sure. Naveen Puttagunta: definition because right now we talking about the platform right now but anyway we are always talking in terms of connecting to a particular data source Salesforce and it for example if you want to remove that the layer of which data source that we are connecting if you want to purely define what is the platform that is holding it again okay for now when you're saying the enterprise brain we are connecting to a particular for example in this context use the salesforce connect to the sales force database. 00:59:55 Naveen Puttagunta: Now bankard brought up the things like how can we bring the SMA knowledge because Salesforce is generic thing right no you're absolutely right come so my thing if I want to remove the the domain specific manufacturing or in no no because that is sorry continue so for example I want to right now I don't want to put any layer to it is a manufacturing healthare doesn't matter but as a platform we have a platform which do these things and if you we can also So on top of this platform we can give the SM knowledge to it to give your output there is some base layer in respect of which domain which industry you are targeting. Now if you're working for in some IRM because we know it's insure type what additional layers that we need to provide on the platform above actually Rakkesh a lot of that domain knowledge comes from LLM itself. Okay, you don't have to teach the LLM about supply chain. The generic supply chain knowledge it comes with that is the advantage of LLM. Okay, you the the generic insurance knowledge it comes with the LLM. 01:01:10 Naveen Puttagunta: Let us say let us start using the term agent. If you say hey I have an agent for salescloud not even Salesforce you have an agent for salescloud then salescloud the agent for salescloud is specifically coming up with information about what is in salescloud a you have accounts uh companies contacts leads you don't have to teach the LLM about what a lead means LLM comes with that knowledge so generic CRM knowledge LLM already comes with you're teaching oh account information is with this entity object how do you touch that object how often is it expected to change how important is it for the for that particular data source uh that is what the agent will publish so when we are developing a new agent for uh Oracle unifier right that is the latest one so the effort that you need to put in it. So eventually when we do the enterprise brain platform standardization and all of that I'm imagining that when you come with the hey we need to develop this new agent for this new uh Oracle unifier it is a matter of days and maximum 2 three weeks which is what you need to do you need to teach uh that agent how to connect to Oracle unifier that is a technical thing that is trivial that will happen then you need to teach it hey generally what are the different entities available in this uh 01:02:44 Naveen Puttagunta: source Okay, contracts, customers, vendors, quotations. What is a quotation? You don't have to teach. LLM knows. But where can quotations be found? How do you retrieve a quotation? That you have to teach that agent. Okay, that teaching and that what are the entities available, how frequently are they expected to change, how important are those entities? uh all of that that agent will that is the work that we have to do when we're building that agent and teaching that agent and that agent will publish that information at runtime. So enterprise brain when you plug that agent in enterprise brain will ask that agent what what do you have and it will publish all of that information. Then the enterprise brain will consume it and says oh then the there is a uh imagine that there is a knowledge collector agent. The knowledge collector agent says, "Okay, looking at that information, looking at its directives for that enterprise, it says, "Oh, my my enterprise admin allows me to refresh data every hour for high priority items. 01:03:50 Naveen Puttagunta: Given that it will go and fetch that data from the agent, add it to its knowledge graph and therefore now the unifier is part of that ecosystem." Got it. Just repeat Nangaru for example if when you connect to any Salesforce for example unifier we are going to fetch the what are all the different entities that are available we are going to define them what that entity mean in our system in in the agent in the agent the in in the agent so based on that LLM has an intelligence to how to interpret the not LLM enterprise brain sorry come come again so LLM has the intelligence to interpret we will call LLM or agent handle both are see LLM is some chart external. Yes. Okay. That is the core infrastructure. Yes. Okay. On top of that we have written enterprise brain. Yes. Plugging into enterprise brain feeding into enterprise brain are all the agents. Okay. For the different first of all when someone asks for enterprise brain we are trying to connect to the the system. 01:04:50 Naveen Puttagunta: That means the agent for example we have a Salesforce agent. Now they're asking us to connect to my Salesforce agent is a request. Then now the definition of the entities what that entity mean in which layer we are going to define at an agent layer or enterprise brain layer. Agent layer. Agent layer. Now for each source each source that corresponding entities uh what they uh what they mean where can they be found? How frequently should they be do they change? Are expected to change? What is the volume? All of that kind of metadata is what that agent will publish. Okay. The definitions are defined on the agent layer. wanted and now because we are also not talking about one particular data source right we are also talking about different one so where we are going to define that this entity is related to another source other entity to interpret the datas so see uh in one way you don't have to okay okay because the what is this agent publishing this Oracle unifier agent is publishing that hey uh Uh these are my vendors. 01:05:59 Naveen Puttagunta: These are you know vendors can be found in this location. Vendors mean people who supply material to uh the company. Okay. Uh vendors can be found here. Vendors are very very you know priority one. They're P1 entities but they change maybe once you know uh at a frequency level order three but uh I orders uh change at a frequency of priority one but orders are priority two entities that kind of metadata it publishes what is an order also it can write but LLM implicitly understands what is an order but even if you define hey orders in this system means that the company is placing an order for uh so many items. Typically an order constitutes of multiple items that also the agent publishes. Now you don't have to connect the dot to if if you it is a supply chain there's another supply chain system that say inventory management system is there. Inventory management system says hey if I'm running low on material for this I might need to place a new order. Oh that order and this order is the same. 01:07:11 Naveen Puttagunta: You don't have to connect because LLM knows that right LLM comes with that core basic information because if you look at you know M policies and all of that you don't have to teach the system what is a policy it implicitly understand that you know it's an insurance domain and it that that industry knowledge it carries that's the advantage we have with the core LLM right so then the agent is publishing how that data source is organized how it is done all of that. So imagine that there is a uh uh data or knowledge agent or collector agent which will look at that what that agent is publishing the unifier agent is publishing and decides how often to pull data from that because agent can't push data. Yeah, this collector agent will pull that data on a you know on a frequency that it is determining based on enterprise preferences and then it updates the knowledge graph right now. So now we have to think of two different things right one is how can enterprise brain be proactive versus reactive. So reactive is I am logging in. So the moment I log in that is also a reactive because I logged in I took an action and then the enterprise brain can determine what is most important to show Naven. 01:08:38 Naveen Puttagunta: Who is Naven? It knows Naven is a CXO. They're generally worried about these kinds of things. If if even if it doesn't know on day one I specified it at one day when I started working with it. I said for me always tell me if there is anything wrong in the company right something like that. So it remembered and it personalizes that that for Navin this is the highest priority. So as soon as I log in it determines what is the highest priority and then gives me that information but do I need to log in? So how do we make it proactive and on you know how can we make it proactive? Because you know will you react for every little um you know action or do you process what is the highest priority for Nav on an hourly basis and if there is something going on send me an alert email or text or SMS or something and say please come to the system there is something that you need to look at or hey something is going on with this customer I can't describe everything that is going on on the email come to the system I'll tell you then it so that is a little bit of proactiveness we We have to see how can we enable that what frequency is it practical or do we we're still proactive but only on login that is something that we we should discuss right how to accomplish that but the core infrastructure is 01:09:59 Naveen Puttagunta: this yeah do we keep a copy of customers data in the platform um Again, do you keep a copy of the customer's data? See, you don't keep the exact copy of whatever is in the database, but you do keep that equivalent embedding or knowledge graph or whatever uh that I got like for example now if you want uh one theoretical one example that I can think of is like if you want to get a accurate number like for example how many projects did wet work that is a statistical data that we want. So there might be a possibility you know a person can ask those kind of questions. Yes. Uh how do we solve those? That is easy right because dynamically enterprise 9 will connect with that agent and ask that agent hey get me the number of projects that ward worked on that agent will find that fetch that data give it to you that is so simple. So I so first of all there are two different things. Okay, the two different things are what is uh what information does the um what kind of knowledge does the enterprise brain have versus what it will ask dynamically to the actual data source and get I want the information that wanker is an employee I mean of course it it knows that because you are a log and all of that. 01:11:44 Naveen Puttagunta: I want the the in the knowledge base. These are some of the things that we have to determine, right? That oh um there is a project called um uh MPI. Okay. Uh it is a it is you know there is a partner called Cortico. There is a customer called IRM. Okay. Then there is an equivalent project for IRM for to develop the IBP portal. So that kind of knowledge graph is essential. Now very specific how many hours have we so far spent on IRM that that level of detail is not required in the knowledge base. That level of detail can be retrieved dynamically from the source system because the agent is always there. The agent knows how to get answers to those questions. And so then if for simply just for uh to move forward what we are saying is uh actually we are not migrating the data. No. So what we are saying is we are you give us the data we don't give us access to the data. 01:12:52 Naveen Puttagunta: Yeah. You give us access to the data. We are building you a layer where we we which is an a first ready uh uh layer that we're building it for you. Yeah. Yes. Yes. Certainly. Absolutely. That is exactly what we're saying. We are not we are not building a data link. In other words, but that yesterday there was a question that Pratima asked. We'll repeat it when she comes back. Right. You know, are we able to do everything that a data lake is able to do? Possibly not. That is not the goal of the enterprise brain. Also the enterprise brain is not a replacement for your data lake. Okay. It it it serves a different purpose. Okay. So Gopal Yashwan you guys Rajashekar G: Yes. Naveen Puttagunta: have been awfully quiet. Any Okay. Rash. Yeah. Rajashekar G: Yes. Naveen Puttagunta: Go for it. Rajashekar G: I have one point here. 01:13:56 Rajashekar G: So as we are saying like uh to answer from the agent needs to answer from my emails or something. So as of now the built system is we are storing the actual data less the embeddings. So the actual data copy needs to be uh uh coming to our enterprise database or whatever wherever we keep but that can be encrypted or decrypted. Naveen Puttagunta: No, no, no. Say, say, say that Rajashekar G: Yes. Whenever we are like uh the fourth column we see like context information emails, Naveen Puttagunta: again. Rajashekar G: meetings, chats, we are not directly accessing the information from Google meet or Google email or something. We are first of all syncing all the information to one database and making them into embeddings and we are using Naveen Puttagunta: No, no, we are not syncing all the information to the database. Rajashekar G: Yeah. Naveen Puttagunta: That is well you may be technically doing that but that is not necessarily what it has to happen is not the thing I I we want to store the embeddings and the knowledge graph but I we don't need to store the actual meeting transcript or actual chat um text or actual email text word for word in a separate database again because the data sources still are the data source are still there. 01:15:18 Naveen Puttagunta: The agent needs to know how to retrieve a specific email on demand. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: Yeah. Yeah. Naveen Puttagunta: Yes. Yeshwanth Reddy Yerraguntla: I want to summarize. Currently, we are is my voice clear? Naveen Puttagunta: Yes. Rajashekar G: Yes. Yes, sir. Yeshwanth Reddy Yerraguntla: Yeah. Currently, we are fetching all the emails and storing it in a column. Naveen Puttagunta: Yes. Yeshwanth Reddy Yerraguntla: But ideally, we should not do that. What we should do is for every employee we should keep track of the source of where that Rajashekar G: Uh Yeshwanth Reddy Yerraguntla: email or chat exists so that can fetch that information on the fly because there's no point in duplicating Gmail and Google chat right now we are allowing that in the interest of development effort and time so this is what Naveen Puttagunta: Clear a shaker. Rajashekar G: one more point here sir. So whenever we need to make it into an embeddings first of all we need to get the data into one place to convert it into vector embeddings. 01:16:19 Rajashekar G: That one time effort needs to be done at any point of time. Yeshwanth Reddy Yerraguntla: Correct. Rajashekar G: After that we can replace those with like we will we will be having a knowledge graph where Yeshwanth Reddy Yerraguntla: Correct. Rajashekar G: is what all the links will be there and embeddings will be with us. All the data sync will be like trashed or something. Yeshwanth Reddy Yerraguntla: Yes. Naveen Puttagunta: Yeah, that you work out with Jes and uh the architect on how to accomplish but that Rajashekar G: Got it. Naveen Puttagunta: but the goal as well as the requirement also please don't try to duplicate all of the email and chat and meeting even for us it it Rajashekar G: Got it. Naveen Puttagunta: doesn't work. I mean it it won't scale because you may be doing that for these what will you do for the enterprise databases you can't say oh I'm going to duplicate all of that data right so you need to there's the same so every data source is the same I mean in in in to some degree that is how we should look at it streaming data first of all you know 01:17:13 Rajashekar G: Got it. Naveen Puttagunta: that's very time based time bound so you may not even have access to the source data anymore more. So you just need to understand the context and store it. That's Rajashekar G: Got it. Naveen Puttagunta: all. Uh uh so now the interesting thing that uh I mean in the real like because we don't have these LLMs and everything so we get the data we try to understand everything. So one point that you mentioned was we have LLMs in hand right now. Uh so we don't need to actually worry about the domain or uh to some degree to some degree. Now then are we saying that you know we are also building a framework even before an enterprise brain what we are going to I mean you're calling it as an agents actually agents or whatever it is. Uh are we saying that we have a framework in place you give us access to your data that framework will take care of uh you have to keep going uh that framework will take care of now I have 10 tables now um unless there is a human who knows that business and look into they can connect to now what we are saying is we're saying that only still we're still saying that we are just saying this framework that we have that we are developing enterprise brain and this agent framework to metadata publish 01:18:50 Naveen Puttagunta: will make it easier for that human to very quickly teach that agent about that data source. So we will develop a new agent for uh let us say uh cautilia. Coutilia is a custom database custom application. It has not more than 20 30 tables. Okay. Now I asked you hey can you bring cautilia into enterprise brain how long should it take because there are only you know see the the underlying connector to the database is a postress connector you have a postgress adapter you know for a better term let's call it a postress adapter that you already have you already have nltosql that capability that agent already has what doesn't it know hey what are the entities in cautilia what do they mean and how are they related to each other. So that I have to teach that agent. By default that agent comes with a scaffolding or in fact even a even a runtime agent because it can connect it will automatically pull the entities out. Let us say it presents to you uh some entities and I say oh yeah this entity is does this entity does this entity does this and I just give a text description of all of those. 01:20:05 Naveen Puttagunta: The agent ingests them and says now I'm ready. So I'm expecting that for a to the scale of something like cautilia you should take no more than two days to develop that agent. One thought process now because in the era do we really need to do that? The reason I'm saying it is maybe one of the business opportunities that we can think of is there are also open source models which means that you can come and they can sit in your environment. you don't need to send your data back to the uh like for example chat GPT if I do a connection automatically the those servers will have all the information that I'm I'm having those models understand what I am what bank is what I'm saying wanker is see cinj database is there there are 30 tables okay it can what I'm saying you are right in that there is a there is a layer that we should also include because for custom databases it can go get the year diagram out you the our agent has let us say has the table it will implicitly understand the connections between them if I have named the tables right potentially it could also understand what the objects mean let us say in my custom application I have not named the tables uh lexically like semantically I have not named they are all abc 1 2 3 You have no clue what they mean. 01:21:34 Naveen Puttagunta: Therefore an therefore a human is needed to teach the system. This is what this means. Otherwise it has no clue what ABCD 1 2 3 4 is. Okay. On the other hand actually because we were saying that you know models will have understanding about I mean not at an organization level but at a high level they understand what is a healthcare domain, what is a pharma domain. Correct. What is I saying? Yeah. So I I just cautilia you have to say first of all what is cautilia? Is there anywhere a description anywhere what cautilia is? No. How will the llm understand what cot is? So we are going to define that. Correct. Yes. Now you are saying hey coutilia tracks resource allocation across projects. Yes. Okay great. That's all you need to say. But where are projects stored in the database? uh because now we are giving that context to LLM. This is a project management based application right I think that it has an intelligence to read those tables and only if the table name is project_ project or projects or project table or something like that if it is named ABCD 1 2 3 4 it cannot determine that that is a project table how you tell I mean I'm open like that is not I'm just I'm actually drilling down Understand? 01:23:00 Naveen Puttagunta: So theoretically what I'm saying is so the column names are the metadata information is not there right but it can understand the values inside it cannot because they they are just numbers for example right that can mean a customer that can mean a project name that can mean anything right so that what I'm saying is the agent should be equipped with being able to retrieve your diagram make those connections and present is this right? To the extent that it is possible, if the tables are named decently enough, if the columns are named decently enough, if the values in the tables are reasonable, then the agent should be able to do 80% or 90% of the leg work. The rest 10% the human has to fill in and say no, that's not what it is. Or if it is in a in a situation like SAP, they are absolutely horribly named. There is absolutely no clue what they are. Okay. So then there there is completely no information about what the column represents. Even if it understands that a particular value let's say VRO information systems is there in in one of our tables. 01:24:12 Naveen Puttagunta: It understands oh that must be a customer right VRO. Okay. But is it a vendor or a customer? Okay. It doesn't know yet. Okay. Yeshwanth Reddy Yerraguntla: Yeah. Uh just one more uh followup example for uh Watana. Naveen Puttagunta: Okay. Yeah. Yeshwanth Reddy Yerraguntla: The Salesforce tables that we are currently using. When I ask about opportunity, there are two tables. One is called opportunity, one is called opportunity_c. Naveen Puttagunta: got Yeshwanth Reddy Yerraguntla: 50% of the time it will go here. 50% it will go there. Naveen Puttagunta: it. Yeshwanth Reddy Yerraguntla: Unless we explain what is what. Naveen Puttagunta: Okay. Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: Right. So the as I'm saying I'm not denying you it should get to a point where the agent framework that we are developing especially for custom databases or for packaged applications it should be able to go and figure it out 80% of at least to the 80% level same du concept based on all the information it should be able to figure out 80% the rest 20% a human tra that is why I'm saying for a application like cautilia the Our commitment to deliver an agent for cautia should be in the order of maximum 2 days because there are only 30 tables. 01:25:27 Naveen Puttagunta: So the teaching in fact I want to be at a level where I don't have to code anything. My teaching will be in the form of a prompt and the system will learn absorb change its own prompt file and say oh that is and it it it updates its own knowledge base or whatever and then the agent is ready. So I will sit with the agent and say uh but so I need an expert on cautilia while your point is why do I need an expert on cautilia only for the last 20%. And in very badly designed databases or applications 50% 60% I might need for SAP absolutely we need somebody for almost 80%. The system will only be able to do 20%. Yeshwanth Reddy Yerraguntla: Is it just me? Rajashekar G: for me Gopal Gottumukkala: Not sure. Not Yeshwanth Reddy Yerraguntla: Okay, you lost Rajashekar G: also. Gopal Gottumukkala: sure. Yeshwanth Reddy Yerraguntla: them. Weirdly, Manishad on the video. Rajashekar G: All right. Gopal Gottumukkala: But maybe that audio might have been hooked up to the Logitech and Logitech wire might have been disconnected. 01:27:08 Gopal Gottumukkala: That's Yeshwanth Reddy Yerraguntla: Yeah. Gopal Gottumukkala: fine. in our thought process what fundamentally we are missing I believe is it is all fuzzy we have been tuned to think binaries for a long long time so when we are working with AI systems also we are forgetting the fundamental that it is fuzzy so we are Yeshwanth Reddy Yerraguntla: Yeah. Gopal Gottumukkala: debating on so there is nothing like a fact in AI world it is an inference even if it says India's president is Narendra Modi is there is some possibility that it Yeshwanth Reddy Yerraguntla: Yeah. Gopal Gottumukkala: might not be. Yeshwanth Reddy Yerraguntla: Good. Gopal Gottumukkala: There is nothing like a fact in this world and the precision rate increases it someday it might reach 200 in certain pieces that is how it should be looked at in Wangut's example Wangut is working on four projects if it goes Yeshwanth Reddy Yerraguntla: H. Gopal Gottumukkala: through a system unless and until it says that you know I pulled out of a database it is not four it need not be four I'm not saying it's not four it is it need not be 01:28:20 Yeshwanth Reddy Yerraguntla: M. Gopal Gottumukkala: four so That level of understanding we have to bifurcate when it is already talking to your system and it is saying that you know I am going to your project management system or I went I pulled up wank and wank is associated to four projects. If it can say that that the source is that and the source says that it is a fact then it could be a fact. In all other places where it went to three different systems and connected everything by inference then it is fuzzy because it made one to n number of decisions in figuring out that. Yeshwanth Reddy Yerraguntla: H. Gopal Gottumukkala: So every decision is a probabilistic uh in nature at a very high level. So it compounds even though the error rate is 0.2%. there is a possibility of compounding error much bigger than what we are anticipating. This is a very fundamental an accepted fundamental of uh intelligence. Yeshwanth Reddy Yerraguntla: Right. Gopal Gottumukkala: What I'm trying to say is when Vasa said something that is his inference of Bharata. 01:29:28 Gopal Gottumukkala: There is no guarantee that that's how it happened. Got it? Right? Yeshwanth Reddy Yerraguntla: Yeah. Gopal Gottumukkala: It came from his mind. Oh, there is no certaintity there. So when you combine imagine Yasa's mind is too Yeshwanth Reddy Yerraguntla: Yeah. Gopal Gottumukkala: powerful and if many many such powerful brains combined together is what we are dealing with right now but still the possibility of uh you know uncertainity still exists unless and until it says I pulled a fact. Yeshwanth Reddy Yerraguntla: Yes. Unless and until it shows the evidence how how it Gopal Gottumukkala: Ah exactly exactly. So there is no inference. Manisha Gundapuneedi: Hello. Yeshwanth Reddy Yerraguntla: fed. Gopal Gottumukkala: inference when there is an inference we should not think like it is a binary system then we'll be discussing about you know why a human is required that's still a debatable business call right if I don't want to have a business I mean human moderation in the picture whatever LLM's capability on that day with the information sources it has that's a fact or possibly a fact is what we what We agreed the industry agreed that we have to keep in mind. 01:30:43 Gopal Gottumukkala: So we cannot claim that it is four. If we are using AI system, we always have to say a system said it is four. Yeshwanth Reddy Yerraguntla: M m. Gopal Gottumukkala: No matter which system it is. Yeshwanth Reddy Yerraguntla: Yeah. Gopal Gottumukkala: So all the layers that we are trying to build up is to reduce that uncertaintity. But no matter how many layers we build, we'll still end up at some I don't know how much some uncertaintity. Did we get the audio back? Manisha Gundapuneedi: It's a Gopal Gottumukkala: Okay. Yeah, we we can control. Yeah, I was what I was trying to say is fundamentally it is a fuzzy Manisha Gundapuneedi: Yeah, got your Gopal Gottumukkala: system. If it is binary, we don't need AI. Binary was solved ages back. So fuzzy is fuzzy. No matter what we do, Manisha Gundapuneedi: Yeah. Gopal Gottumukkala: nobody else can come suddenly and make a fuzzy into binary system. Fundamentally dies. That's where we came from. Right? Binaries are limited. 01:31:46 Gopal Gottumukkala: So the decisions are supposed to be fuzzy. Heart is heart. It changes. Your heart is different. My heart is different. Everybody's heart is different. That that in that fuzziness we are trying to infer the data. The data is fact. So Wangard worked on four projects is a fact. If we went to cautilia if it went anything more than cautilia then it suddenly becomes fuzzy. Manisha Gundapuneedi: Got Gopal Gottumukkala: So that we have to keep in mind. Manisha Gundapuneedi: it. Gopal Gottumukkala: So if we want to add a moderator, we'll add a moderator. If we don't want to add a moderator, we don't want to add a moderator. That is a business call. That business call could have been taken by an agent or business call could have been taken by architect. could be anything. If you bring in AI into that kind of decisioning also it goes to the next level but we we can never attach binary into that entire go anywhere then that that's a limitation again we are inherently killing the possibility of getting it better. 01:32:50 Gopal Gottumukkala: So in another way if you wrote an if else and the possibility of improvement is dead there itself you understood right if I wrote an if else where I'm supposed to write a switch statement then I killed the extensibility the same principle here it's just an analogy but the in our minds we should not have that it is a binary system we always have to remember it is fuzzy entire solution that we are trying to uh build is also a fuzzy system. We cannot convert underlying I am using so many language models and I cannot say I made it deterministic that is not possible. Manisha Gundapuneedi: Go ahead. Gopal Gottumukkala: So I was only trying to solve a problem. I mean address one point that you are discussing with Navin and LLM already knows right that binary statement is dangerous. Manisha Gundapuneedi: Yeah, Gopal Gottumukkala: LLM can only understand to the level it has. It is like you and me. How much you understand Kilia, how much I understand Koutilia is drastically different. What Navi knows is totally different. Manisha Gundapuneedi: good. 01:34:07 Gopal Gottumukkala: We we any one of us even today Naven also cannot say because Naven doesn't know how much information and what kind of information we are pushing with how much of precision that he defined though Naven invented the tool that is a state right so even today if Naven sits and teaches an agent it cannot be 100%. again because Naven might have taught everything. LLM has fuzziness in it. There's no way Naven says X LLM understood as X is a theory but not binary. There is nothing like yes and no here. It is always a possibility of yes and a possibility of no. So when we are designing thinking all the places it has to be that Manisha Gundapuneedi: Come Gopal Gottumukkala: way. you have to leave that probabilistic uh um end Manisha Gundapuneedi: on. Gopal Gottumukkala: outcome accepted. If I am anticipating I built a a first dashboard and if I'm anticipating that the structure is supposed to be exactly X then there is no point even working towards it. It cannot be X. There is always something that comes in because underlying data sources that we are connecting are changing right every day they're changing. 01:35:26 Gopal Gottumukkala: Suddenly Vro could become you know something else. Classic example is Java became Oracle. No more Sun exists. Manisha Gundapuneedi: Yes, Gopal Gottumukkala: Classic example. Then if I ask who is the owner of Java. Manisha Gundapuneedi: you Gopal Gottumukkala: If that data has some reference that Java was owned by Sun and acquired by Oracle, there is no way an LLM can figure out that it was Sun. You getting the point? Manisha Gundapuneedi: got it. Yes. Gopal Gottumukkala: So it all depends on what it has at that point of time and keep on how much it is getting further. Manisha Gundapuneedi: Yeah. Gopal Gottumukkala: So what Naven was trying to talk about that agent is not a deterministic deterministic agent then we would have called it as connector. What he is trying to say is it interprets whatever it can depending on what model that we hooked up to understand what that source database image video doesn't matter anything. Got it right. Somebody has to accept the level of knowledge it has is reasonable for us. I mean whoever that consumer is as of today we don't have another system that can also be mechanized. 01:36:36 Gopal Gottumukkala: So Naven brought in a human moderator into the picture just to close the loop otherwise Manisha Gundapuneedi: All Gopal Gottumukkala: immediate question comes is how do you know what it understood? We hooked up a pretty standard Oracle ERP zero customization. Still the understanding of the database by a very well tuned LLM less than 100%. Mathematically right. Manisha Gundapuneedi: right. Gopal Gottumukkala: So Naven is closing that question there itself. So when it requires there is a moderation. So if somebody will come and moderate or not that doesn't matter. But in the design there is a way to make it precise if somebody wants to. Manisha Gundapuneedi: Correct. Gopal Gottumukkala: Yeah. So N did I put that aentic? Uh uh okay. Manisha Gundapuneedi: Correct. Correct. Yeah. Gopal Gottumukkala: Yeah. Manisha Gundapuneedi: The other thing that the minor logistical element that I was also implicitly including is we have to build the agents in such a way that uh you know all of that training all of that customization whatever you do is kind of contained in the agent right and so dynamically that agent can just plug into the enterprise brain framework right so we'll have multiple agents and if an agent is upgraded then You upgrade that agent with the upgraded knowledge or upgraded this thing that it will publish and so 01:38:07 Manisha Gundapuneedi: dynamically it can be consumed on by the platform and so we have a very uh dynamic uh way that you plug in agents and then agent self-publishes. So the teaching is not all centralized in the platform. You can independently upgrade the different agents, right? And so you can publish different multiple agents and so on, right? And so it it is a scalable framework rather than you have to say, oh, first you so that is the difference between a agent and a connector for me. Connector is dumb. So agent is somewhat intelligent that it it carries with it some metadata, some information, some learnings that can be incorporated into the enterprise grade. Yes. That's does that work? Yeshwanth Reddy Yerraguntla: Yeah. No. Manisha Gundapuneedi: Okay. So, so we we need so the take away from this is we will rearrange this information in a way that we are talking about hey here are your data sources here is how we categorize the data sources and okay and then the other things are all there right so we will not say information consolidation because immediately they will uh you know the discussion will fall into what is the difference between this and data lake we will say we will build we are trying to build an enterprise knowledge graph, an enterprise knowledge based knowledge graph from multiple sources of information consolidation immediately we say why 01:39:53 Manisha Gundapuneedi: are you I mean because intelligence is what see we are trying to make an intelligent layer in fact for me it is more about you're trying to rebuild a human being here. In other words, right, there is a nervous system, there is a um uh there are sensory organs. So in my head, enterprise brain, it is not just a brain, it is the because it is a nervous system, right? You're trying to connect different sources. There is inputs from different places. The output can I mean we are saying AI first and in AI first thing the output can be a graph it can be a video it can I mean whatever right so yeah it is more about re redoing the anatomy of a human being so okay second that I keep feeling. Gopal Gottumukkala: That's Manisha Gundapuneedi: Sure. Gopal Gottumukkala: one. Manisha Gundapuneedi: Because empathy comes from the heart. I I feel that as I Gopal Gottumukkala: Yeah. Manisha Gundapuneedi: sure Gopal Gottumukkala: Navin just before we move on Wenut many parts that you dis you you you mentioned while discussing with Naven are still valid uh Wenut LLM has knowledge we are not denying it but the only thing is it's just fuzzy and then uh you mentioned something else uh why do you need to send something to LLM we are just saying LLM we are not saying local or cloud it is just an LLM which has knowledge as a model learned with something and that's what we are referring as LLM. 01:41:47 Gopal Gottumukkala: We are nowhere saying that local, remote, personal, you know, shared that is all you know the decisions can be made at any point of time that doesn't make any difference in the construction of the solution as long as LLM is there. Okay. So yeah those call could be very dynamic and it all depends flow to floor. Somebody has an ecosystem which is rich enough that they can they can borrow an LLM or you know clone an LLM put it into their own ecosystem it's their but others Manisha Gundapuneedi: Uh yeah. Gopal Gottumukkala: are Manisha Gundapuneedi: So the reason I brought it up and yeah at the highest level it looks like you know it doesn't it doesn't make a difference whether we are hosting it. I mean whether we're going with open source or thing like chat or anything uh but let's say for example if you are uh going with an LM which is open source. Now we also I mean when it comes to implementation whatever solution that we are planning to implementation there will be a work I mean there is going to be a work that is associated with how do we I mean how do we host that model of the finetunings and everything so that will become an uh aspect to consider that is why I brought up that point and 01:43:11 Gopal Gottumukkala: Yeah. Yeah. Absolutely. Manisha Gundapuneedi: the Gopal Gottumukkala: You are giving options then we have to work. There's no question on it. But we should not be thinking that why we need to go to cloud. That should not be a question. So only thing is if we are accepting local LLMs, we are assuming that that LLM is is very equivalent to any other LLM in all other aspects. That just an ass assumption. This work has to be done. Manisha Gundapuneedi: Yeah. Okay, next. So this we have already discussed that uh these are going to be the landing page or dashboard whatever but these are the parameters that most likely somebody at the se level would look for. Actually this also we need to reook at it and I would uh what I would say is um Rakkesh this is your space. So what are we trying to say here and what do you are you thinking about breathma? So okay first Gopal Gottumukkala: Um and PMA the uh if I understood the intent behind this 01:44:34 Manisha Gundapuneedi: go Gopal Gottumukkala: presentation while we are designing the system anything and everything that you would have asked earlier plus what you could not get answers for enterprise brain will solve These type of sentences we need on what aspects of that capacity we want to present when we are presenting a solution but this should not be driving force to design the system or I mean design the solution it should not matter actually whether it is region specific or anything the solution should be there is knowledge you ask the question so these things should come how I want to suppose I wanted to sell I assume that I have an enterprise brain I want to sell the enterprise brain to Navin what I would like to project to him so that he'll connect with me and then I can explain the solution but these statements should not be the driving factors towards the solution that's what I that's what my perception Please Manisha Gundapuneedi: This This is a sample categorization or classification of information that somebody would want to look for. But if you want to give a different perspective, 01:45:54 Gopal Gottumukkala: I understand sir no no no no no sir sir I am not saying these are Manisha Gundapuneedi: that's also fine. Gopal Gottumukkala: finalized I know that these are samples my point is even if they are Manisha Gundapuneedi: Yeah. Gopal Gottumukkala: samples they these samples or even the real ones should not drive the solution solution is about no I came to Naven I'll ask a That is how the solution should Manisha Gundapuneedi: So, so there are two aspects here. Gopal Gottumukkala: be. Manisha Gundapuneedi: pocket when it is an when I I come I ask a question what the solution or what is it that the the application is going to return is one thing but we are trying to build an AI first where we are trying to build an autonomous system autonomous system uh where the system is driving information to me continuously So the the way so I'll break this problem into two parts. One is classification of the information or the classification of the categories and this is how we have designed it so far and whether we like it or not most of the B2B platforms they bro the classification into it. 01:47:20 Manisha Gundapuneedi: I'm challenging this team to figure out what is that uh holistic classification that feels that kind of represents our thought process. Uh when we are thinking or when you're solving you don't break things down into update separately growth separately. So okay end of the day I'm saying that the way we think is a combination of thoughts that come hit us and then out of that we figure out we come up with inference and we do something right that thoughts can be that thought right it goes from what we can do to what is a risk what is possible what is not possible it's a combinational so If brain human I want us to reook at how we are classifying let's challenge ourselves to move away from the traditional classification this is one aspect that I'm talking about the second aspect that I'm talking about is based on the question one of the structures I could think of was like this descriptive insights diagnostic pred prescript this this this one because we end of the day while all of while the system can be as dynamic as possible we also for us to find solutions and to come up with patterns we need structuring around this right so that is where I am coming from so that layer may work when I'm asking a question is there something beyond it let's challenge ourselves But if it is AI first where the system is driving information to the user continuously then how would I do that? 01:49:31 Manisha Gundapuneedi: What what considering the data sources considering the agents how would I do the uh in uh the wisdom and decision acceleration on the screen. So I have two aspects here. So see my intent is not to question my intent is not to say one of the key things that I've been observing is if we don't challenge ourselves right now we will go in a different direction or we have already gone in a different direction. So we need to correct the course right now is how I felt um because Manisha Rasher for that matter even Rakkesh and me right if given a choice we would our comfort zone is to think what we have done in the past so are they thinking from this so I'm trying to figure out how else to do this so and I'm not saying what we are putting up is wrong. Even in the top one from earlier slide, my intent is not to say email is wrong or chat is wrong. My intent is to say we need to go to the depth of the problem in every aspect and challenge ourselves because otherwise this is what any other person is doing. 01:50:59 Manisha Gundapuneedi: Everybody is building LLM solutions. Everybody is building platforms. Everybody has products, right? And everybody is doing the same thing. That is the current challenge right now where end of the day front dashboards are looking what they used to be with some little bit of changes. So I'm literally challenging ourselves to start going the other way and figuring out can we can we can we think of this like a human brain and figure out as a human how do I think and reclassify all of this. I think I repeated the same statement multiple times knowingly because for me that uh is needed right now. Yeshwanth Reddy Yerraguntla: Can I anchor the discussion with one sentence that you mentioned? Okay. Manisha Gundapuneedi: Yeah ma what is Yeshwanth Reddy Yerraguntla: You said, Manisha Gundapuneedi: it? Yeshwanth Reddy Yerraguntla: "What is it that causes me to take decisions quickly?" I will rephrase it in a way. What is it that makes me take a decision in the most effortless manner Manisha Gundapuneedi: Yes. Yeshwanth Reddy Yerraguntla: that I am confident about what I consumed. 01:52:11 Yeshwanth Reddy Yerraguntla: I'm confident about what I processed in my head. And I'm confident about what I'm about to decide and why I'm going to decide. If we can think in these four five lines, what does it take for me to get that confidence? Maybe that will help us understand what should be on the screen to uh be in such a state. Manisha Gundapuneedi: Can you rephrase and I'm I I understood at a high level I understood but there is there is still a a lot to understand probably to understand the depth of your statements like for example Yeshwanth Reddy Yerraguntla: Yeah. Manisha Gundapuneedi: Right. I I just took a break and somewhere I was reading something and it it said visual clutter adds 40% to cognitive load. Yeshwanth Reddy Yerraguntla: H correct. Manisha Gundapuneedi: Okay. And if cognitive load is high, decision making power reduces. Yeshwanth Reddy Yerraguntla: Got Manisha Gundapuneedi: Okay. So I was trying to infer. Yeshwanth Reddy Yerraguntla: it. Manisha Gundapuneedi: So generally UI visual clutter you should know what is a visual clutter visual disturbance versus visual enhancement and this is a statement I also use that it increases cognitive load but I never was able to put it like how that cognitive load is going to hamper my decision making or consumption of data right so I'm with How do we classify? 01:53:55 Manisha Gundapuneedi: Maybe we need a discussion around this generally. How does anybody has any idea about um I mean anybody read anytime about how a brain functions because I was reading about something something recently just now I was reading there's something in the okay there was MMT something in between brain which which needs to continuously um it needs to be fed with boring stuff and for it to be become stronger and stronger. So monks in Bhutan or somewhere or all the monks right morning they get up and they do the most boring stuff which is cleaning the entire monastery. Y and T that kind of builds the resilience of a person, determination of a person. And so the more boring stuff you do, the more u uh uh willpower increases and that's one I mean I was just reading that and that strengthens something in the brain. MT G. So yeah basically I'm try I was in context of enterprise brain I was trying to understand what is this brain how is how how is our brain broken down so for example um IoT information what would you relate it to um SAP information how would you relate it to the empathetic side of human uh how would you relate it to left brain right brain um when a certain part of the brain goes down right they say that it can't do certain stuff then what is that so I was just trying to relate it like that I mean I don't know if it is too complicated or I'm I'm I'm thinking uh very differently 01:56:06 Manisha Gundapuneedi: but um then what's the point of calling this brain if it doesn't work like a brain is how I felt about Gopal Gottumukkala: Uh if we are calling Manisha Gundapuneedi: But I'm definitely challenging. Gopal Gottumukkala: something yeah now if we are calling Manisha Gundapuneedi: Sorry, Gopal. Gopal Gottumukkala: it as a brain yes it should work like a brain but the problem right now we have is what is a brain itself is thousand different definitions exist. We have to figure out one of those definitions and stick to it. And all the thousand are not far away. Manisha Gundapuneedi: Yeah. Gopal Gottumukkala: Fundamentals are same. It is exact dynamics. The books differ. Manisha Gundapuneedi: Yeah. Gopal Gottumukkala: But all the books talk about the same fundamental about how neurons work, what they store, what they put it back, when they can bring it back. Those theories are fundamentally more or less very very similar. It's just that exact dynamics when they bring it into the level of mathematics then people have different paths but fundamentals more or less everybody agreed as of today at a very fundamental level brain doesn't want to remember anything like something is this it doesn't want to remember the fact anything and everything it would like to process and bring it out because facts cause memory memory. 01:57:33 Gopal Gottumukkala: So brain doesn't have that much of memory. Manisha Gundapuneedi: Yeah. Gopal Gottumukkala: It only inferences can be put in and then again it will it is that is the reason embedding scale uh not I'm not being very precise and it cannot be very precise. That was the very fundamental theory of uh uh what I know Manisha Gundapuneedi: All right. Gopal Gottumukkala: to the limited knowledge that I have. It was a huge uh um breakthrough that it is just inference. It is inference and it is not memory brought in the artificial there is a huge uh you know improvements came because of that one understanding it's commonly agreed by many scientists and then of course lateral many many things came we have to remember those when we are building these solutions that's the point that I was talking to bank earlier also keep that fluidity and we will know something more tomorrow and that could change our course also to that level. We have to be flexible in our thinking, solutioning, building. We have to keep constantly having that which is a very fundamental of how brain works. 01:58:48 Gopal Gottumukkala: But again it is like you know whether Rama is a person or a god it is something very similar to that. Yeah. I mean if to answer your question uh has anyone read about I cannot say yes because I never thought what I'm reading has solidified anything for me. It just gave me few few few hints um and I constru I constructed a logical understanding of it and it is connecting to the way I am thinking and that is constantly changing even how brain works understanding is also changing day by day. Maybe that is how it is intended to be but I don't know. Uh it could be totally wrong. Manisha Gundapuneedi: What the Gopal Gottumukkala: Also earlier you mentioned about two things we have to understand how we want to Manisha Gundapuneedi: f***? Gopal Gottumukkala: present in in the second way that you mentioned right. Is it contextual or is it structural or is it some other way that we have to obviously keep in mind when we are solutioning because no matter how our brain is going to send information to us there we have to make logical calls again that could also be another decision engine but ultimately we will only have handful of representations that could be five or 50 but it is very deterministic in the sense we only have n number of ways. 02:00:16 Gopal Gottumukkala: N is very uh there is a boundary to the n and only that is the number of uh ways that we can pass on the information to the user. So we have to keep that in mind and there is a destining system and we are going to be deterministic at a certain point of time that is very well valid. Yeah, others I lost a bit here and there. Uh that we have to keep in mind. What I was saying is does it support growth related or is it to be a primary or is it to be a vertical that this model should be able to do growth uh related information in my mind it could be dangerous because I'm putting verticals too early into the Manisha Gundapuneedi: again. And last Gopal Gottumukkala: no u the other one one things that you mentioned like you Manisha Gundapuneedi: point Gopal Gottumukkala: know are we getting executive updates are getting growth related information that should not drive uh you are thinking. Manisha Gundapuneedi: Yeah. Gopal Gottumukkala: So yeah those two are I was stuck at those 02:01:26 Manisha Gundapuneedi: Andamentally at an architecture level what we are trying to put there. Um and again we get all our inputs through the five sensory organs. uh somebody uh and spiritually somebody was saying there are 5 + 7 + 7 I don't know 14 plus 19 spiritual organs and that was one of the events anyway we'll stick to five but five spiritual five sensory organs you get you get uh inputs now inputs we have a conscious and a subconscious mind in the background which you think of it as a belief system temporary storage permanent storage whatever. Now the external input and what we already have from in our head or mind together make us think a certain way, feel a certain way, react a certain way. Okay. How we think and how we feel is what makes us react. The reaction is decided in the brain and it is thrown to one of the organs like hand, leg, mouth, eyes these. So there are input organ input stimuli there are external outputs internally and then there is a nervous system which constantly feeds something else signals and all of it. 02:03:19 Manisha Gundapuneedi: So I was trying to figure out okay how will I map all of this to the enterprise brain because end of the day what the the knowledge that Wenard was talking about or Nin you were mentioning this is all for me building a conscious mind and a subconscious mind. Subconscious mind, conscious mind is the one that is getting active and then the brain is the one that is receiving. So this is how I was looking at it. But that is exactly what we trying to do with enterprise brain. Right? In other Gopal Gottumukkala: Yeah, Manisha Gundapuneedi: words, Gopal Gottumukkala: most of the energy I could able to interpret and uh connect and but yeah in a way yes but we We have to make it a little more precise. Manisha Gundapuneedi: yeah, I I didn't know how to make it any um I was and these are the thoughts that come, but I never sat and solidified And the reason I'm bringing up all of this is to challenge all ourselves not to fall into the trap of regular thinking. 02:05:07 Manisha Gundapuneedi: Okay, we go. We have not even completed the first part. Yeah. Yeah. When you put that on, I smiled because that is the objective. So you go through and then based on what you consume each department will individually brainstorm on what they wanted to present to accomplish the goal and then we come up post lunch. I mean that is the goal. All right. What are the next steps? Huh? Gopal Gottumukkala: Hey Manisha Gundapuneedi: Uh, okay. Gopal Gottumukkala: everyone. Manisha Gundapuneedi: So, so design team can come up with new parameters. This workshop is for that design come up with how I look at is you wanted to present it this way and engineering will say how they have to do which technology that they have to do and what AI can accomplish. Uh I'm with you. I would and that workshop in the in the in Gopal Gottumukkala: Thank Manisha Gundapuneedi: the spirit of the workshop and I would say that uh Gopal Gottumukkala: you. Manisha Gundapuneedi: um we should do that together collaborative. 02:07:06 Manisha Gundapuneedi: It's okay that it is serial. No, that's fine. In fact engineering we don't even have to be part of this workshop where okay E requ if we identify that you know what will it what is it to be done how it should be presented things like that then we can independently go figure out what is the right engineering architecture AI how we'll support it all of that the workshop let's Yeshwanth Reddy Yerraguntla: Yeah. Manisha Gundapuneedi: do it together not breaking up. I know when you presented it I didn't want to interrupt at the time but I would say that see just like this right even for us to get clarity on what is AI first system it took us a while what is AI first UI it took us a while only on that collaboration it it can be done this kind of breakdown of how information should be looked at then refinement of that we should only do that in in you know that is the intent of the workshop I would say that let's do that together. Fine. So in the spreadsheet itself, can we have one more sheet for this and then we kind of fill in? 02:08:15 Manisha Gundapuneedi: Absolutely. So what what are we talking about? What are we trying to say? Gopal Gottumukkala: What Manisha Gundapuneedi: So if we are assuming that or this is our assumption, Gopal Gottumukkala: else? Manisha Gundapuneedi: what are better put it as we don't want to go this way as a conventional application. We wanted to redefine it and then challenge. So what are the parameters that we are thinking? Uh-huh. What are are we talking about business parameters that unconventional way whatever you're saying? See, what is the first thought that you got? Not you. What did you get? That's Rajashekar G: Yes, in my context like we have already tried something on this. Manisha Gundapuneedi: good. Rajashekar G: So like last week we connected we need to build a simulator kind of thing. So how a dashboard a personalized dashboard should look like for a CX or CEO someone. So when I see this what I have seen is uh if Navenaru logged into the system. So there will be one block uh there will be some KPI saying uh this is your quarterly target. 02:09:38 Rajashekar G: This is the number you have achieved. That is the first one. And second one these are all the VIP emails you have got. Uh these need your attention. You need to reply to this like that. And those kind of things came into my mind whenever I see this kind of a thing. Manisha Gundapuneedi: What about you? hotel. Maybe what is the thought? Um but so the reason I was asking you all is because we we train ourselves to think a certain way dashboard. We trained ourselves to think about it in a certain way. Okay. And I feel existing dashboards have extreme cognitive load and they are not decision enabled. They are information enabled and probably wisdom enabled but necessarily not decision enabled. Right? Yeshwanth Reddy Yerraguntla: H.B. Manisha Gundapuneedi: I mean by default this because naturally we think like that that is leading me back to the 2D dashboards that and those dashboards have huge cognitive load. Okay. And this is one thing that I constantly keep asking everyone. 02:11:28 Manisha Gundapuneedi: Look four. What what does four mean? Nobody knows. I know because the assumption that 2D dashboard is making is I know it is four compared to 10 maybe in my head. Oh with the with her or her background and right. So Adi what we are doing is the inference of the information on the dashboard is heavily dependent upon my past information and lot of processing and then only a decisions are made. Okay. This is this is a challenge with the existing dashboards. Now decisioncentric dashboards is one concept we have talked about even in the thing but next project but this is not the cognitive load is extremely high. Now how will system because for example analytical data numbers my cognitive load to numbers is or let me say cognitive load I have good cognitive load my tolerance to numbers is very low I don't even Usually this is me. He's never happy. He wants you to explain the entirety of it. Right? So he you call it cognitive load, you call it tolerance, whatever it is, right? 02:13:35 Manisha Gundapuneedi: My tolerance to financial numbers is very low. His tolerance to financial numbers is extremely high. So Naven should be presented analytics. I should be presented summary right. So this AI first dashboard should understand all of that and give me a ni one way. Second way, how we present the data should be decisioncentric, not information or wisdomcentric. That is another one I'm looking at. And what is the just right dynamism creates a lot of uncertaintity. There should be certain pattern because as humans we all like that pattern. There is a pattern to our life but within each of those blocks. So any blocks science. So I don't know how to present this data. So if but if you're thinking them as levers probably they may be different. Enterprise brain might be consuming and giving me the pathway is different. The front end consuming and showing it could be different. So in my head there's a lot of complexity involved here. There is a thinking layer there and like the the there is a thinking layer then there is an intelligence layer on top of the thinking which drives the direction. 02:15:45 Manisha Gundapuneedi: Then there is an emotional feeling layer. Then there is subconsciously there is something else which is driving all of this. And then there is an output reactive layer. philosophically abstract but if unless we think slightly differently we won't get a a dynamic decisioncentric dashboard which kind of builds the trust of humans maybe we may not come up with a solution may not but I need us to think beyond the regular way that's where I I'm Yeshwanth Reddy Yerraguntla: aka two questions. Manisha Gundapuneedi: coming. Yeshwanth Reddy Yerraguntla: First is uh um what does it mean to have a decision centric uh I don't know view like okay example on maybe we'll have more control on how to think that is one input I have second input I have is uh kpa are bad now these uh traditional way of showing things are bad on number but why are they bad like what is the problem if I continue to show them like no it's not like the outside world is uh really reluctant to take this anyway if I give them and say okay today I go to this customer and say you know what we have this thing called enterprise brain um it has all these uh Lego bricks of ways of showing stuff to you. 02:17:36 Yeshwanth Reddy Yerraguntla: Um, and it also the best part of it is um, you decide how you want to set it up. Um, eventually you can always personalize it in the way you want just by talking to the system. You don't need to do even clicks or drags or anything. It's not complicated. Somehow we'll nail it. We'll nail it and we go to the customer and say okay end of the day you're still seeing KPIs. You're still seeing graph charts. You're still seeing uh whatever paragraphs of text. These are the conventional things that can be shown as part of a normal dashboard. Will they will the customer really be reluctant to accept it? What is the danger in continuing with the normal you know traditional or safe way of figuring out these are the normal things to show anyway? Let's continue doing that. Right? What is the danger in that? Manisha Gundapuneedi: you will not there is no danger there. 02:18:30 Manisha Gundapuneedi: uh on the same note I have a follow-up question. I'll take it whatever Ashant has put it in. We just take an example like for example I'm looking at a dashboard which is talking about uh uh sales happen at a region level. So so before that uh sorry um are you when you say you're connecting that question to yours uh are you agreeing with him and add on question or I'm agreeing with him. Okay. So uh so like when you says you mentioned one important aspect decisioncentric dashboard like for example if I'm looking at the sales data across the region it will tell me that you know across this continent you know the sales is good this continent the sales is not good now anyway in a traditional way I can drill down and I can find out uh you know why that is diagnostics you're not decision you're not leading me to decisions there is still information in my That is wisdom and you Yeshwanth Reddy Yerraguntla: Sometimes that's sufficient. Manisha Gundapuneedi: are comparing Yeshwanth Reddy Yerraguntla: I'm saying let it be. 02:19:33 Manisha Gundapuneedi: sorry Yeshwanth Reddy Yerraguntla: Sometimes isn't that sufficient to to know where I stand? Okay, sir. Manisha Gundapuneedi: you're asking me but I'm ward is calling it a decision I'm saying it's not decision it's still information okay because there are it's still information Yeshwanth Reddy Yerraguntla: Okay. Manisha Gundapuneedi: I have to consume. My brain has to still consume. Then I have to understand so there is the the lag between information to a decision. There is time that I'm taking to process all of this. It is not decisioncentric. It is informationentric. So okay now on the other hand uh I mean if the system says uh uh this because of these these reasons u you know the sales are down and these are the actions that it has to take now do we consider it as a decisioncentric uh as long as you are giving me quick decisions to take yes it is decisionentric now I'll come to your question but first I'll answer yeshwan's question yesan I mean are you see there is nothing wrong it this work. 02:20:43 Manisha Gundapuneedi: I mean that is how the world is behaving right now. But you tell me how many of you guys in this room use Yeshwanth Reddy Yerraguntla: Yeah, Manisha Gundapuneedi: dashboards for consumption. Yeshwanth Reddy Yerraguntla: I don't Manisha Gundapuneedi: You don't you do by reading not on Yeshwanth Reddy Yerraguntla: Mhm. Manisha Gundapuneedi: dashboards? No, you don't. Right. Yeah. I'm trying to tell you guys I have to consume a lot of dashboards. Yeshwanth Reddy Yerraguntla: What? Manisha Gundapuneedi: Okay. Okay. I need information from every department. Yeshwanth Reddy Yerraguntla: Yeah. Manisha Gundapuneedi: I need to consume a lot of dashboards. I hate all the dashboards. Okay. Yeshwanth Reddy Yerraguntla: Okay. Manisha Gundapuneedi: I I literally look for one person every I mean I look for an executive assistance who is smarter than me. Literally movies. Uh I don't know how many of you watched um Madame Secretary or any of these movies. I know you watch but others where a executive is walking and the lift opens or the door opens. Okay. 02:21:49 Manisha Gundapuneedi: There are three four people behind them trying to constantly feed them with information and sometimes summary for them to make quicker decisions. In fact they even probe you to take certain decisions. Yeshwanth Reddy Yerraguntla: Yeah. Manisha Gundapuneedi: They're understanding that is how see when you are when you are saying this is an enterprise brain and if it is being designed for the CXO layer of enterprise brain CXOs they they they race against time okay every small thing is important. Now when these four people who come to the lift and start feeding this person with those information and decisions right they're the most trusted people. So blind they believe in they just blindly trust with everything that they can just believe in whatever they are saying and start taking decisions. You get my point? So now to answer your question yes BI Yeshwanth Reddy Yerraguntla: Yeah. Manisha Gundapuneedi: platform or Tableau or for that matter Rajett was not the last meeting but the previous meeting how that Oracle unifier gives a dashboard gives a dashboard all the softwares independently give lot of information okay weekly 02:23:26 Yeshwanth Reddy Yerraguntla: H. Manisha Gundapuneedi: So these people Yeshwanth Reddy Yerraguntla: Mhm. Manisha Gundapuneedi: secretary they are much much much smarter than Narendra Modi is you're Yeshwanth Reddy Yerraguntla: H Yeah, I'm Manisha Gundapuneedi: understanding a that is Yeshwanth Reddy Yerraguntla: falling. Manisha Gundapuneedi: what is important. So now going the traditional way does it harm? No, it doesn't harm. But what's the point in building this enterprise brain when we are not disrupting the market? But we we need to distr If you're doing it, might as well disrupt the market, right? Disrupt the way people think. Yeshwanth Reddy Yerraguntla: Got Manisha Gundapuneedi: No. Now go, Yeshwanth Reddy Yerraguntla: it. Manisha Gundapuneedi: right? Why not take humanity to the next level? Save my time. Yeshwanth Reddy Yerraguntla: Makes sense. Yes. Manisha Gundapuneedi: Okay, I answered your question. Now your question if I don't trust the platform they're like my executive assistant if I don't trust my executive assistant first of all I mean sometimes uh I tell Rahul and I'm in a mood you see you you figure out what is going to come your way either I'm hungry you are not hungry do you need to order a break that's what executive assistants do they know exactly every damn detail about a person. 02:25:13 Manisha Gundapuneedi: If 2 minutes late and they read that human to that degree and you should watch all of these time whenever you get a chance like man secretary there some when she's walking in some days they bring her a scone because a morning some news so she's stressed out they know that the scone will relax her. So this coffee so there are there are many many things that are done to pacify you to for you to take better decisions also when you're calmer so multiple layers so information but information Okay fine applications applications and everybody is doing integration everybody is consolidating and giving you information again then what is how are how is diwami different design with design and AI design and AI we understand humans we is calling itentric platform then it is notentric platform in my head Yeshwanth Reddy Yerraguntla: Got it. Manisha Gundapuneedi: final point right I just want to understand what if I trust that person right or if I trust that context other require I'll then ask for details okay now what why are you suggesting this when I don't trust when I need information or when I think the way that representing is slightly different from what I'm expecting. 02:27:07 Manisha Gundapuneedi: You guys are understanding. So then I get into details if required. Otherwise if you just give me oneliner I want to go with that oneliner. Why the hell should I break my head about me identifying the oneliner 3 minutes but what's the last line first? What are you saying? I the question here is how do you think? So the I mean the answer I was expecting is I'm thinking it in blocks of information like a traditional dashboard. That's all. Then if I ask why then you can explain right if we all can stick to this each one of us will save each one of us at least 2 hours a day minimum. So enterprise brain in intention is to reduce the time to decision for whoever is using it because time is the most critical aspect in the current era. What if you don't make a decision now and if you change shift that decision to finement I don't know what anthropy is going to release that is going to make or break me I don't know what tsunami is going to come that is going to make or break me so time is the most crucial thing right now in this 02:28:49 Yeshwanth Reddy Yerraguntla: Yeah, that helps a so I can also start thinking now what does it mean to what does it mean when we say we need to surface information that needs quick decision uh taking capa capacity. So I am thinking like evidence I'm thinking to last line Manisha Gundapuneedi: There are Yeshwanth Reddy Yerraguntla: first recommended actions the evidence of what is happening why it is Manisha Gundapuneedi: last Yeshwanth Reddy Yerraguntla: happening these are things that probably can drive 80% of the time because most like if I think about it when I need to take a decision I need like probably six or seven things um that I have to keep in mind while I'm taking that decision that that takes us back to what Gopasa was saying right uh on average human mind can only store only few things it's not a uh memory machine it's a machine that wants to do processing based on information Manisha Gundapuneedi: Got it. Yeshwanth Reddy Yerraguntla: so to surface those six seven things whatever those six seven are for that decision taking step I think should drive like a lot Friction. 02:30:20 Manisha Gundapuneedi: Yeah, think of it like that. And truth be told, works like about 20 hours a day. Even then he cannot follow one of the things that I'm dead against in Diwami is recording of meetings and then saying here is the sharing of the recording of the meeting. I really hate rather than I recording the meeting I just say tell me the gist of it and I don't why should I waste another half an hour 45 minutes going through the entire recording right it's useless the time is very critical so the point that I'm trying to make is there are five or 10 people feeding him with lot of information and where they are summarizing it, putting it in an intellectual form for him to be able to consume at his cognitive load and also suggesting him how he should use that information. But the final decision lies with Narendra Modier. But these guys influence his style of thinking, what he feels and how he should react. Um followup question but then you know when we're talking about enterprises what we are saying is proactively we are going to do do present that information um and it is not hard to replicate anybody can do it. 02:32:02 Manisha Gundapuneedi: So what is the uh selling point or why why should people come and uh look for enterprise brand? What we are doing is we just presenting that what are the decisions that they can take anybody can build it actually. Yeshwanth Reddy Yerraguntla: That is Manisha Gundapuneedi: No, that is that's the most difficult there Yeshwanth Reddy Yerraguntla: Oh, Manisha Gundapuneedi: is I mean not decision but it will say needs attention for example I always tell you there is a one there is if there is a problem there are thousand solutions and downtown picking the right solution based on the context and the audience, who you are talking to, what you're talking, everything matters. See, it is not white and gray. There's a it's not like, okay, let me put it this way. There are facts and truth. Truth cannot be changed. Like is a truth. It's a fact. We will give the same answer. It's the truth and fact. truth and facts key there is there is no perceived uh in interpretation of stuff but anything beyond truth and fact everybody looks at the the information with their perception okay the way you perceive information and way I perceive the information are very very different right and the way we want to present it is also very very different right the perception of information matters the most and picking the right solution or the decision in the context of that perception matters the most like I'm saying there could be thousand 02:34:10 Manisha Gundapuneedi: decisions right They're very superficial with they are they are very very very my that's what I was before we came to this slide this is what I was talking about our brain has many parts this enterprise brain that we are building You should also have so many parts. Now what these parts are fine of it. I also want us to build a subconscious mind and a conscious mind. My subconscious mind is nothing but the beliefs that I have. Your subconscious mind is the beliefs that you have. They are very very different. Right? The conscious mind is the one that is taking the decisions. Right? And me decision will be different. and our decision will be different in the same context. Okay? We will get it like for example we want to go to dinner. Okay. If we are not if the context the decision is which restaurant to go to for dinner right the context there is our time low if if I am home and he is home he will say somewhere here oh and fancy it's no okay Now my decision will be co his decision will be oake he does not like Japanese I love Japanese I don't like Punjabi he loves so who now who is the deciding factor and decision everybody knows who is the 02:36:21 Manisha Gundapuneedi: decision you understood all of these factors are the ones who enable based on what data. So enterprise brain should be suggesting to pratima go to kokai probably to nav go to now it should also know hey nav if you're smart and if you're going with pratima just pick kokai oh is in a good mood she rested all day she will be okay going to oake then you should say Yeshwanth Reddy Yerraguntla: Make Manisha Gundapuneedi: hey sgestake she's fine today something along those lines that is question like I hope you're understanding it we are building an intelligence that is the intelligence information intelligence are two different things and continuous I'm saying existing dashboards are information information whether we have to use our intelligence and then take a decision with AI in place it can build the intelligence to what degree I don't know that is where I need yeshwan and these guys you guys have to Over a period of time we can build a huge intelligence to begin with in intelligence. Something on those lines you guys should tell I mean coming up with U patterns is not a big deal right trying to understand how we should consume all of this. 02:38:14 Manisha Gundapuneedi: What is it enterprise brain is going to throw to the UI and stay then how to present it is what I was looking at I'm not saying I won't do this anything is fine just to clarify we are not actually asking for give me UI pattern that is the last thing that I want sorry please I don't want to get into that is not what I meant was saying something I was I was talking to aren't you? Yeshwanth Reddy Yerraguntla: and cut now. Manisha Gundapuneedi: I am saying I actually want Can I say what I want? I want What kind of stuff will you want AI to throw? AI can throw whatever you want. We're we're going in this circle. A can throw whatever you want. What kind of stuff do you want to AI to throw? Then we can imagine how to make it happen because a can do Yeshwanth Reddy Yerraguntla: Yeah. Manisha Gundapuneedi: anything. We will figure out how to make it happen. What kind of stuff should we if you give us examples of hey this is the kind of stuff to throw. 02:39:26 Manisha Gundapuneedi: Then we'll figure out oh what will be the underlying architecture to enable all of that? How to think about oh now today they're asking this but tomorrow they might ask this. So those kinds of ideas will only come once I have an concrete example. That is what I'm asking. Pick one use case. Let's walk through that. What kind of stuff you want AI to throw? Then we'll figure out how to throw that. Yeah. Yeshwanth Reddy Yerraguntla: uh yeah like I wanted to ask Watana if there are so many things out there that support decision making. I wanted to know if like what are those things? Manisha Gundapuneedi: No my no I'm not okay let me rephrase my sentence Yeshwanth Reddy Yerraguntla: Mhm. Manisha Gundapuneedi: Ashman so if decision making is what we are uh giving to the user what I'm saying is there are I mean that can easily replicable what is the USP of our whatever we are building as an enterprise brand right how is it going to be different from other systems is what I'm 02:40:31 Yeshwanth Reddy Yerraguntla: That is where I'm curious about. Manisha Gundapuneedi: saying and and I also Yeshwanth Reddy Yerraguntla: No, no, that is where I'm curious about if they are easily replicable and if they are out there, Manisha Gundapuneedi: mention Yeshwanth Reddy Yerraguntla: I want to know what are they and I want to see them how they are enabling decision making. At least as a starting point, at least as part what we are trying to go against. Manisha Gundapuneedi: Sorry. Yeshwanth Reddy Yerraguntla: If you can give me examples, Manisha Gundapuneedi: Whatever decisions that we saying right at least to me to to me it feels like you know everybody can do Yeshwanth Reddy Yerraguntla: give me one example. Manisha Gundapuneedi: it. Yeshwanth Reddy Yerraguntla: I'm not challenging in any way. I just want Manisha Gundapuneedi: I can also build that. Correct. Uh then how is it different from other if if person like me can think that I can build then how are dash Yeshwanth Reddy Yerraguntla: It took us like 3 hours with all of us sitting in one place. Manisha Gundapuneedi: different Yeshwanth Reddy Yerraguntla: It we started to realize decision making and time saving or what are needed. 02:41:45 Yeshwanth Reddy Yerraguntla: Is it that trivial know I'm asking myself actually why didn't I think of this if it was so trivial. Manisha Gundapuneedi: So on one hand wanker let me answer you it is um it is not so whatever we are talking about hey this is how we should you know we are thinking we have to enable decision making we're essentially building uh you know decision intelligence systems that will help humans make faster faster and better decisions. Faster and better decisions. We are also trying to scale human intelligence. That is um you know there are only four uh you know let us say there are seven of us. Sorry to interrupt. Think of it like a calculator. Okay. I mean generation at least think of phone number. So basically in day when when something else can do your work we move on to other things and ma yeah right that is uh human evolution. Yeah sorry please go ahead. You know that was that was just a point I was trying to say. 02:43:17 Manisha Gundapuneedi: Go ahead. I lost the channel. You are talking about decision making and you are saying that think of it like this acceleration of decision uh decision uh we have to and those are the key factors and that wasn't the point I was trying to make the what we're building is a you know faster and better it'll help humans make faster and better decisions right so it's a decision intelligence system the Second aspect is we're also AI will also help scale human intelligence. That is the way to position autonomous automation system. Why do you build automation? Because you are saying hey a human is not needed here. It's not that a human is absolutely not needed. The human was needed earlier to determine what the flow is and now you're scaling it up. You're making it you know thousand humans. You're you're automating it right what would have otherwise required 20 humans you so you're actually scaling human underlying it is actually human intelligence that you're replicating right so those are the that is the kind of way to think about it now we are trying to build these better this intelligence systems and we're saying hey this is one approach to do it in in to enterprise brain will enable this this this and this it is not so trivial that oh everyone in the world right like you know can actually think through this 02:44:56 Manisha Gundapuneedi: whole approach uh and then build it to the scale and level that we are we are able to productionize it we're able to build it we we put our heads together we we build better user experience better you know this thing all of it we we are trying to do right imagine that we are able to accomplish that like with UX all of that that is not so trivial that every company is able to do it but it is not so unique that we are the only ones in the world who will come up with this okay for example Gemini is in this path Gemini wants to do this that was the original thing that we started with you know uh what is his name uh my seven I'm Um Google that's money money his last name is money they are so they what they wanted is they wanted the adapters or connectors to 200 enterprise uh applications. So they're already on their way. They are they are the ultimate right because they are going to connect each and every enterprise system. 02:46:23 Manisha Gundapuneedi: Okay. And so they are on that way. So they're going they are building this. It's not that we so because they already have hooks into Google Drive, email, chat, everything, right? They're going to do it just like them. Microsoft is going to do it. So everybody is going to do that. The reason that I'm asking this question actually on Monday's call we had with he was talking about what is the differentiation factor. So I'm just trying to understand or bring it to the point of discussion. What is the differentiation factor that we are going to have with so we have to take a positioning approach. Okay. What is our differentiation factor? If you try to go up against Gemini for an enterprise, you will not win. There is no differentiation. There may may not be a differentiation that will matter to the enterprise. We might still claim our differentiation saying that we have better user experience. We are rethinking the whole because at the end of the day you know how are you able to consume this you know some company might come up with enterprise win and their only end user interface might be a chat window I ask questions you would is that the best user experience no because it's only solving a reactive problem right I have to ask the question if I don't even think about asking the question the right way then it will not answer so some companies might be going 02:47:44 Manisha Gundapuneedi: that another company might be going another way where so in the end of the day what actually is a differentiator the user experience is a differentiator right now in the enterprise against Google would that be a differentiator against Gemini user experience may still be a differentiator but the enterprise needs connectors into 200 systems you don't have that so it will not be a differentiator so what will matter what so we have to pick the positioning right so we have to pick a Our positioning is okay are you a uh midsize company midsize what do we call mid-market company right mid-market company by definition they don't have a IT huge IT team they have IT operations team but they may not have a development team so okay they can pick Google Gemini but Gemini will cover some aspects of it are they can they pick the other databases where where do they get from the other databases. So we are able to plug in. In fact, we are not saying we're competing against Google. If Gemini is providing some part of it, we will plug into Gemini. 02:48:53 Manisha Gundapuneedi: We will use Gemini agent, right? We will we will provide a better user experience for the domain for the specific problem statement that that mid-market company wants. We will build a custom solution for them. That is where our services will come in. Right? So you may for a SMB we probably are a much better offering or maybe Gemini might be a better offering for them because they may not have even custom databases and all of that and they may be able to live with whatever user experience that they're getting from Gemini. So that way so we it the differentiation will also come in positioning. So it is not an absolute differentiation is never an absolute because you may have a differentiation but if the customer doesn't care about that factor that is no longer a differentiation right so that is why you that is where we play actually sorry just to complain Yashwant was making some point before he forget what was that chart can you put up the chat sorry everyday ma Before I forget, one activity that might come handy for all of us could be to retrospect at the end of every day in the next one week to see which decisions Ma could have empowered us with at the beginning of that day to have made that day more productive for us in an ideal scenario. 02:50:18 Manisha Gundapuneedi: I'll tell you I I mean this is something that I really really need if you can make my life easy. But before that ward the differentiator that Dwami is going to bring this we are a design company. All right. We claim and we should that we understand human psychology and human behaviors far more better than all the other companies. We know how to present the decisions much better. when the minute you thought decision making you probably I'm assuming you must have envisioned it some form what you vision and I am envisioning so it's perceived value I'm envisioning that you visioned like a bunch of uh statements so in your head it feels like right No, I'm not saying that we'll only present it like that. You know, only drama. There is drama associated with it, right? Every agent comes with an aura. That's why I was saying you know how these people are received how they are received there's drama around it drama but end of the day with all that drama crucial decision see acceleration is all about decisions right you anything that you sit on because you're not able to decide you know there There's only two or three possibilities if you put your head to it maximum right it is our nonacceptance because of which we just sit on it and don't take a decision you're 02:52:16 Manisha Gundapuneedi: understanding you're understanding the how a human we understand that really Well, so I am saying wisdom into possible decisions are decisions nudging him and how important that decision is to take right now is what will accelerate the business. Okay. So yes, if you are asking me everyday morning, what I need is what are my lowhanging fruits? What are the most value fruits in a sense? one crit what are the highest value that is where I want and then you have to focus. Yeah. Sorry. Go ahead. No no no just on that note if Prabhakar or Prabakar or Raja remember that is what I was talking about is that I wanted enterprise brain immediately to be deployed into Diwami so that at our level Pratima especially I was taking Pratima's example we will understand what are the kinds of things that we are able to get and then we will reimagine the whole enterprise brain how to present it because until there is a practicality everything looks abstract okay and fundamentally if you think about our environment in Dwami actually 90% of our enterprise knowledge is embedded in email and chat and Jira okay only very little we because we don't depend on our uh you know enterprise applications as much everything is done in uh you know in these days now we have moved to WhatsApp also right so the core of that enterprise knowledge is there so the source is not important because the whole concept of the agent framework is you are demystifying that source you have a new source two days we'll give you that 02:54:49 Manisha Gundapuneedi: agent the the core is in the platform where it is able to absorb that knowledge, learn new techniques, know learn new skills, remember them and then uh replay and then so the real real real differentiator is how are you able to present it to the user. We are struggling so much sitting in this room to figure out what makes sense to the user, how to display and all of that. Okay. And we have been in the user experience business for 17 years. do not trivialize that we are actually very good at imagining user experiences and we are struggling to come up with this new because enterprise brain is so undefinable. So that is where the differentiation is end of the day everybody is able to build that is the you know musk and all these guys are saying that is the lesson that is why the the whole IT industry fell 20%. is because cloud work, cloud code anybody can build. It is the experience that you can bring on top of it. That is the differentiation. 02:55:56 Manisha Gundapuneedi: Okay. Now everything has context right. If you are going against enterprises where the primary value for them is you know connecting 200 data sources they may not value they will value it second but first is that so that is why positioning is also important. Yes. Yeah. Yeah. I know it's very late for you. 2:30. 2 1:30. Yeah. Sounds good. Yes. Thank you. Personalized on the person who is using it. Yeah, that is one thing and I wanted to run something by Pratima and Rakkesh. See, generally what have we been taught in user experience? Hey, good user experience there is predictability. Don't surprise the user. You know that learnable learning curve is very high in the UI and change management is low. If when we say that the whole UI is is dynamically changing it is based on context context is that that cognitive load is my learnability that are you compl that that seems to be conflicting with that first principle that you know don't surprise the user it has to be predictable the learning curve has to be low and all of that what do you think no it is a concept of the container I mean for ease of understanding for everyone the number of containers have to be fixed or the size of the containers or the type of containers have to be fixed. 02:58:11 Manisha Gundapuneedi: What comes into the container can be extremely dynamic because end of the day everybody needs a ground to stand on. Now ground can be very different. There should be some consistency in the screen. It could be as simple as a logo and then uh navigation on the left side, logo on the right side and then uh break the dashboard into two parts. Something there should be some ground which is consistent and there should be there should be a lot of dynamism within them. I hope you guys understood what I'm trying to say. People want certaintity that certaintity definition can change quite a bit. The certaintity could be there is a left bar navigation on the side. The certaintity could be when you click for drill down it'll show up only in a panel. The certaintity could be the logo will always be there. The certaintity could be the colors or the background. The dynamism of the data and the content can change to what degree an it all depends upon who you are presenting it to and what their cognitive load is and how they will perceive it. 02:59:21 Manisha Gundapuneedi: Now end of the day if these dashboards can build flyovers for them to accelerate their decisions first time nervous after that it'll be fine it is like um Harry Potter or Hogwart stairs they keep changing but first time but once you are once but stairs are still stairs you know that the stair height is that much the stair is going to be there. The stairs keeps moving where it connects what it doesn't connect is dynamism. So mentally you prepare them for you know what are the constants and what are the variables. End of the day people want some variables. Sorry people want some constants. A constant percentage can be 5% 10% 20% depending upon the tolerance of the user how anxious they feel or usually CXOs typically tend to be a lot more dynamic than a person in the bottom of the tree they want lot more constants they cannot take a lot of dynamism so the human behaviors come into picture what because CXOs deal with dynamism every second, every minute. So but what are the constants right? 03:00:45 Manisha Gundapuneedi: So, so Harry Potter met is the right example for so we we'll also take a break and go Gopal Gopal Raj we'll take a break we'll just uh we can still do a working lunch but it may not work okay we we'll keep the discussion Gopal Gottumukkala: Sure. Manisha Gundapuneedi: open yeah yeah so take a break we'll we'll we'll be eating and then discussing you also grab something. Rajashekar G: Okay, sir. Manisha Gundapuneedi: Okay, keep it open. Oh, I see. So, What time is I just want to get Charges that sarcasm. 11:30 11:30 11 record. Volume fish think the size. Session ended after 03:05:27 03:37:09 Manisha Gundapuneedi: ally being automated because whatever you're executing I will automate you know AI can do it right you if you are if you generally build spreadsheets that is automated if you are you know you're not a decision maker you are just an executor so executor will be automated where decision making comes only there this one will help so let let us take a middle middle management suppose you are a project manager, right? 03:37:42 Manisha Gundapuneedi: Like you're a delivery manager. So you you also take a lot of decisions. So then this kind of system will be your you know chief of staff. So we'll be so okay it is because your role is let us say so focused on project execution, project management. So then you know uh the Jira plug-in Jira Jira agent is is really what will drive. So it might say uh because you have also visibility into the sales pipeline. It will say hey the sales guy is going to talk to uh you know they have a meeting with your project stakeholder for a new opportunity on Tuesday next Tuesday. Okay. Um you have a deliverable on Monday. If there is anything that happens to that deliverable that talk doesn't will not go well. So you better deliver it by you know Sunday so that Monday you have a call and then put them in a good mood right. So now what is the decision that you need to make? Hey you know maybe it knows that there is looks like there is a risk that one task is going out of uh out of the sprint you you might miss it. 03:38:55 Manisha Gundapuneedi: So do you think that you will you know what are the what are the possible choices? You can ask uh you know these two people to work for three more hours or you can you know in this other uh project most of the milestones are done. There are a couple of people with unallocated uh hours there. Maybe you can ask to borrow them to uh get this thing done. Okay. Or the third one is deliver a simpler. I mean this one may not be even possible for the AI to figure out but you can say you know there there is an uh there is a this one is a higher level task that is going out of the sprint but there is a much much lower prioritized task that is in the scope. maybe you can switch the order so that there's no risk of the high priority one falling out because that has a much higher impact on the sales discussion. So therefore this is the decision that you need to take. Okay, how do I either I borrow more resources or I ask one more person to so you think about uh you know maybe if the AI is around it can also do the repercussions analysis right okay these people you can see they have they have taken lot of vacation over last month so 03:40:15 Manisha Gundapuneedi: therefore it is you know perhaps it is okay to ask them to you know ask additional or you can say oh these people have not taken vacation in the last 6 months maybe you don't want to ask them to spend the weekend, right? And so a better decision might be to ask the other two people to come join. But uh because they we see that from history, they have already worked on this project. So they kind of know this project already, right? That level of that is the thing that you usually you know if prabhakar prabhaka will go and ask the squad lead you know what do you think and then the squad lead will suggest over these things. So that is what we're getting right. So for the pure task executor there is no value to uh this thing because first of all that role itself is going to be eliminated because the you know ex everything pure execution is going to be Satyasri Prabhakar Mantripragada: Wow. Manisha Gundapuneedi: automated. That is why again AI has two primary roles right scale human intelligence that is the scaling part where because at the pure task executor level that human intelligence has been automated so you're able to scale that human intelligence. 03:41:36 Manisha Gundapuneedi: So then the the next step of decision better decision decision intelligence you are providing that is what we are providing. Satyasri Prabhakar Mantripragada: Got it. So one point scenario um not denying your scenario but wanted to understand it Manisha Gundapuneedi: Yeah. Satyasri Prabhakar Mantripragada: better uh in a scenario for example you were mentioning that uh as a delivery manager you wanted to track the project progress at the same time enterprise brain indicates you that there is a possible meeting with the same client ongoing today uh for a prospect or something whatever by some other team. So in that case one maybe the delivery manager is not copied on it but still enterprise brain is exposing that information. You're not supposed to know because that's a different team that is handling for a different context not related to this. There could also be a possibility that it's just a regular followup call on how things are working on the project side. How are you happy or not? So in that case should we expose that information to somebody also part of the project? 03:42:41 Satyasri Prabhakar Mantripragada: How do we define those governance rules Manisha Gundapuneedi: Yeah, I I am not getting into because to me that is purely a logistical issue. Satyasri Prabhakar Mantripragada: there? Manisha Gundapuneedi: Prabhakar, I'm not getting into that. See um at some point what will happen? First of all, Prabakar the uh sorry if I can you want to finish off your thought? Go ahead. Pra the way we are executing things right now in Diwami that is becoming the benchmark for all our discussions. I want us to knock that off first. Satyasri Prabhakar Mantripragada: What the f***? Manisha Gundapuneedi: Second thing the way all of this will be done. I mean I I need all of us to understand and accept that how we behave, how we act, our processes, everything is going to get impacted by this AI layer. So the use cases and the context that you are all bringing up as examples in our situ in in our today's discussion are based on the past and I'm repeating again and again our behavioral patterns our reactions the way we think everything is going to change in is going to drastically change in the next two next three months to one year two years if we don't change we will fall behind as simple as that for all you know maybe in future there won't be any meetings maybe AI will ensure uh you know and the autonomous uh autonomousity of all of this 03:44:30 Manisha Gundapuneedi: may lead you to only have strategic meetings maybe nobody is interested in your status calls anymore. Maybe maybe the way we take coils will change. So a lot a lot is going to change Praa. Satyasri Prabhakar Mantripragada: uh in general not only concerning about but in general your VP sales Manisha Gundapuneedi: So Satyasri Prabhakar Mantripragada: it's some other part of the world and the development it's at some other part of the world project manager sits at a different part so when they working on a project based on the timeline. Apart from the current deliverment, there's some escalation content that was posted on a enterprise forum and uh the same guy is not going for a follow up on the same product but they wanted to crossell other products within the ecosystem. In that case, if somebody says, "Oh, I already am already using this particular product of yours, but uh the progress on this has been very slow. Now, how can I go with this new product of yours?" So, those are the scenarios that I have 03:45:42 Manisha Gundapuneedi: Correct prabagar. Satyasri Prabhakar Mantripragada: seen. Manisha Gundapuneedi: To me right prabhakar that is purely a logistical issue that will get sorted out because right now the enterprise data is so siloed people don't automatically give because they think that okay somebody dealing with tech support system um they don't need access to the sales uh thing because their role is tech support why should they have access to the sales pipeline but the tech support manager manager at least at the you know level of that manager needs to know if there is a sales order coming in. Generally this is where a lot of that collaboration comes in. In fact, in enterprise systems in enterprises, when the sales guy is going after a company that they already have with, they actually pull this meeting together with all the internal stakeholders, you know, delivery manager, tech support manager say, hey, I'm going for a a meeting with them. Do you have any uh escalation? Do you have any issues? All of that, right? So the if they forget to do that that is when they get into trouble because they go face the customer when there has been an escalation and the customer blast them and then this guy is all pissed off he'll come and say why didn't I know about it but he didn't ask the question so it's nobody I mean so then it it gets fingerpointing and all of that that happens all the time in enterprises right so but that that is a logistical issue because the nobody uh people didn't say that oh tech support 03:47:17 Manisha Gundapuneedi: manager needs access to the sales. Even if he has access, he's not going to check continuously, you know, if there's a deal going on and all of that. That is where enterprise brain will have to come in. It has to correlate that distinct pieces of information and help the user make a better decision, right? It has to help the salesperson say, "Look, there is something that you will be asked about." you have a meeting with that customer. There is something that they're going to ask about. It is going to let the tech support manager know that hey there is a sales you know there is a sales meeting currently there scheduled for today. So most likely the latest issue that got escalated will get highlighted in the sales thing and will have an adverse impact. So then it will help the tech support manager do something different or they make the right decision. Say here are your options. You accelerate your solving and then report that the problem has solved before the meeting or second you let the salesperson know. 03:48:19 Manisha Gundapuneedi: But why we discussing certain scenarios like these uh to I mean what is it that we are trying to get out of these scenario discussions? He asked a question to he asked are we exposing different pieces of information where how will we take care of governance I'm saying they will get sorted out those are logistical issues they will get sorted out there will always be a governance issue where you have to have a governor in place that will ensure that you are not exposed to information that you are not private to that is always going to be so Gopal everyone is talking about security AI governance then AI autonomousity then AI intelligence where and as far as I am understanding governance is an additional layer that is being built I'm inferring it from the articles like that so for enterprise brain also do we need that governance layer and if so what will go into Gopal Gottumukkala: Yes. Governance comes into three different levels. If if it is AI system, there are three levels of governance. Basically, we we have to have the monitoring built in. So whatever is coming out of monitoring, the system should have a capability to take decisions at different different levels. 03:49:53 Gopal Gottumukkala: At an infrastructure, we'll capture the telemetry and present and then uh infra will make calls. At a model level, we'll keep looking at what the model is doing against what it is supposed to do. And then an architect has to capture and then there should be some interface or some other mechanism uh should be able to decide to continue to do the same thing or is there something that needs to be done? Governance is not not just about uh watching and leaving. It is about controlling it also. Something is not supposed to happen but an alert come then go back and then fix it. that that same Manisha Gundapuneedi: And Gopal Gottumukkala: thing. Manisha Gundapuneedi: theoretically enterprise Gopal Gottumukkala: Yeah, I do then. Manisha Gundapuneedi: Then Manisha that is one thing. So basically right before one of the biggest uh for lack of a better word in assume that we have been an experimental mode where now I think we need to get into alpha mode or beta mode however you want to call. So specifically okay system what are the things that are required what are the different layers that are required from the architectural point of view is one layer what should agents come agents different types of agents empathetic agents agents what What are the different kinds of agents that are required in the enterprise 03:51:55 Manisha Gundapuneedi: game architecture? Maybe I you guys have to sit with Ashwan and Gopal uh and come up with the proper architecture and different systems what security mean I don't know how much you both have. Rajashekar G: Sure, ma'am. Got it. Manisha Gundapuneedi: Go approve action items. Satyasri Prabhakar Mantripragada: For the current system that is in place, Manisha Gundapuneedi: Ask Satyasri Prabhakar Mantripragada: we have we are implementing the basic governance at least user level we are protecting that whatever information the user is entitled to will only be visible to the user even when we query for the external information. So for example within Dwami uh because that is the current scope that we have implemented so far. If we roll out enterprise doi then you and I both will log into the system using our own credentials for Google log. Uh still I cannot ask any information related to you. uh those uh security measures are taken. Gopal Gottumukkala: Raak that is a functional feature. Satyasri Prabhakar Mantripragada: But there are other Gopal Gottumukkala: So in general governance will fall into multiple layers. 03:53:20 Satyasri Prabhakar Mantripragada: players. Gopal Gottumukkala: So we were talking about that. So what you're talking about is a feature that an enterprise green needs not as an a system governance or an enterprisegrade product governance. That is just one functionality where we call it as governance. Manisha Gundapuneedi: Guys, guys, I'm I request all of you please participate. Let's close it and go and then do whatever. Gopal Gottumukkala: Hello. Satyasri Prabhakar Mantripragada: Whatever we thinking. Gopal Gottumukkala: Okay. Satyasri Prabhakar Mantripragada: You went Gopal Gottumukkala: uh but yeah that is also you know Satyasri Prabhakar Mantripragada: down. Gopal Gottumukkala: conceptually a governance but that's a feature that we built in system low indulion governance is not that is not the governance so what we were discussing earlier is in general if you are building uh an AI system what kind of governance features that we should embed It is slightly different. That's what I'm trying to say. Satyasri Prabhakar Mantripragada: What could be the examples? Gopal Gottumukkala: So whatever the governance that you are talking about right that is inherently part of every system. There are users and there are roles then it has to be controlled. 03:55:12 Gopal Gottumukkala: In general any system that we take in general without going into the business logic of the system what governance means is what kind of level of control that we have on that system when it is going beyond the way it is supposed to do in the in the sense um configuration also comes into it. system configuration monitoring and alerting comes into it uh comes into it. And I'm not talking about the inapp notifications. I'm not talking about those. Whatever is infra, it is supposed to be in certain certain shape. How do you know that it is in shape? Satyasri Prabhakar Mantripragada: Oh, Gopal Gottumukkala: I I thought two nodes will be enough. But is a system running with two nodes in a happy scenario? In sense, is it happily running or is it struggling? Is it time to put the third node? So every aspect should have exposed how it is doing the its job excluding the business. Then do we have a any mechanism built in to take corrective measures. 03:56:26 Gopal Gottumukkala: We don't need to build a system to automatically correct some aspects. Yes, some aspects most of the aspects are provide the ability to the people who are going to maintain that application to look into that application and make sure it is running in a good shape. Satyasri Prabhakar Mantripragada: Huh? Gopal Gottumukkala: That aspect is we call it as Satyasri Prabhakar Mantripragada: What? Gopal Gottumukkala: governance. Say how many 401 errors came into the application and then did we get an alert? Joshua get an alert saying that these many errors came. If it is 500s, did Va get an alert and did Va has a facility to see that number and drill down into the number and then figure out an area where those are happening and is it a pattern or is it happen only today? That kind of ability is application should have governance features. and we say that that that's what we usually build while we are building any system. So in AI there could be other uh control points that we need to pull out and then provide an administrator or maintenance people so that it runs in the way it is supposed to run. 03:57:46 Gopal Gottumukkala: A system could bring too many too many other things. Everything else was we know what it is supposed to do and if it is not happening then we know but in here that is much more dynamic. So there are many other uh telemetry that we need to pull out. Hello. Satyasri Prabhakar Mantripragada: Get it in. Got it. Gopal Gottumukkala: Yeah. Whatever we built or we are we are building conceptually that is also by English it is a governor. So we named it as governor but it is a business feature that we are building in that application. Hello Manisha Gundapuneedi: And okay. So what is uh next step? Can we uh are we at a point to progress to maybe discussing how do we want to do this workshop that you know discuss a use case and see what would be the you know uh what would come from AI and therefore what we could um how you would structure UI is that what we want to do how do we proceed what what is the next step. 03:59:21 Gopal Gottumukkala: I have a thought that I wanted to correct myself or uh say yes or no to think in that direction or not. No. Say I take the same uh is it okay if I uh continue? Okay. Manisha Gundapuneedi: Please continue. Gopal Gottumukkala: So earlier Wenut spoke about an example of uh how a certain business aspect is doing across geography. He took an example of sales numbers across I mean uh categorized by continent. Imagine somehow without beating the UX principles I could able to project look into the Africa. I I represented Africa. Manisha Gundapuneedi: You could be able to look into Okay. Gopal Gottumukkala: Context could be anything. Just wanka took an example of sales numbers. It could be two different ways depending on the business application where the sales is happening too high or sales is happening Manisha Gundapuneedi: Yes. Gopal Gottumukkala: too low. Then I just immediately thought okay what I will do is pattern doesn't change. Africa will come wherever it is coming but it will be of a little different size than the remaining six continents at a very lowest level. 04:00:36 Gopal Gottumukkala: I am helping the user to make a call because he's supposed to look into it. If I extend my thought process in that I would have provided some more hint into the same card. This happened because of so and so or some hint already which used to come after two drill downs. Now I'm bringing up and saying that so and so happened because of this. You better look into it. The moment I touch that I already helped the user what he supposed to do. Manisha Gundapuneedi: Hey, Gopal Gottumukkala: He did it. He clicked and then went in there. Then I again went into much much deeper and said that okay talk to this guy and then reduce something. I mean uh reduce the price, Manisha Gundapuneedi: Yeah. Gopal Gottumukkala: increase the uh increase the what do we call this offer price something like that. Is that I am not saying that is how the UI is. I am just saying is that the thinking right? Manisha Gundapuneedi: We'll take seven. See 04:01:54 Gopal Gottumukkala: This is Manisha Gundapuneedi: scenarios are very clear. There's no ambiguity in the scenario. What are the cases and all that? But the only thing right because there will be multiple scenarios to a single persona. How can we expose the scenarios to to that person and still if you are wanting to particular scenario allowing you to drill down to get more context of it. These are all it's like individual problems for example in traditional things right because we don't know we tell that we understand that persona well we know what kind of information is needed and we put that everything in a different widgets and layouts and all that is what we exposed to them later also they said that no my priorities keep changing quarterly. So we said that okay we can give a configurable widgets you decide what you want to go so we have given that and we and still we are not able to call. So in traditional what we call need attention section where your focus has to be don't need to worry about different widgets and all it's come into the picture now still the thinking of design is still there now only thing right we don't want to go with the traditional dashboard because if I bring those things it's nothing you won't get any difference but even the intelligent layer is there in the background the users won't get the feel of it's an intelligent layer 04:03:10 Manisha Gundapuneedi: because they're looking the same layout same widgets and all that you know we are settling what is the pattern that is going to break that's is the struggle otherwise the cases are very clear the approaches are very clear the reason we are not able to move forward from here so wink do you see where the differentiation could be if we are struggling so much on user experience you know and we are reasonably good at design and you see how critical AI and design are interlin right without design understanding how to expose AI it will just fall down to chatbot or you ask questions I'll answer and and that is where the differentiation is while everybody can build an agent everybody can build a Salesforce connector what is the big deal you can do it in two days these guys did it in three two days everyone else also can do it in two days the core idea of that email you know collecting data that these are not new at all these are things there are products out there that are trying to do it NLSQL there's a startup called wisdom.ai AI that their only job is that NL2SQL so just to do the the work that we have done because we not we don't want to go with a problem directly so what we have done is we try to put our thoughts like a 04:04:41 Manisha Gundapuneedi: scientific approach to understand what are the different archetypes because it's like our domain is very big personas are also varied so we try to put them into an bigger bucket called archetypes so uh what the personality that defines the archetype so some are like very fast and action oriented like the core pattern is like a speed and momentum. So like that we put we don't want to forget too many also we created only 10 to understand what is their uh cognitive load in understanding the information and their what their pattern and momentum to work on it and what is their trust because the other thing is like how can we establish trust to those kind of a persons. So what kind of elements that we need to bring onto the table to bring the trust experience to the user and what the enterprise brain patterns would be for that personas. Now after that because we believe that this is 10 for a bigger gamut like multiple personas and multiple roles that you're playing that won't be sufficient. So what we have done is like under the each archetype we try to put them into an different persona. 04:05:43 Manisha Gundapuneedi: The multiple personas coming into the same archetype it's not shared. share my yeah so even though they the rapid decision Gopal Gottumukkala: There it is. Manisha Gundapuneedi: maker they itself there is a way even the the few people want an evidence to see before to act on certain things so we put them into a different personas that are falling into thing and each person has their again code trait what is the conclusion first for them some are like they want to see the more information before they act on anything they have some frictions we need to clear those frictions before they take the actions so these are all the Different patterns are defined here. Now we what we have done is like entire organization we break into three dimensions. Now the one is like visionaries and executives. The other one is the managers and the operationals and individual contributors ICT. Now how we are telling that this archetype this persona which one is a dominant for them this personality or a trait and everything is for executive dominant or it's a manager dominant. 04:06:42 Manisha Gundapuneedi: So that will system will have an intelligence to show what kind of a pattern is needed for them and what kind of then explosion. Then we try to understand because in order to make someone someone who newly logged into the system where to put which archetype and which in the archetype which person it belongs to at the beginning of the application you may not have all the attributes to consider but you have some basic attributes among 30 attributes you have only five attributes. Based on that maybe you put into some archetype some persona bucket but based on the patterns that is going to work on the interacting with the system based on the the deep uh he's going to get some data or conclusion that he want to arrive system will have a mobility mechanism the mobility it can be same archetype moving to a different persona or the mobility completely shifting the person from one archetype to another archetype. So for what are the triggers for them to switch those things is like a learning for the system to do. Now we have a clarity on what are the archetypes and what are the personas and what are their patterns what they prefer under the system. 04:07:44 Manisha Gundapuneedi: Now we went to understand the information what are the different types of information that come into the system. It can be there also we try to put an archetype layer. So similarly we have an entitycentric that means always talking about a things are attribute. It's always a fact. So like a key or a tulle or a record there is no it's a true fact and there are the state ccentric depends upon the state and scenarios keep changing and there are few timecentric based on the time and series it changes like so here also we try to bring that kind of an archetype behavior and what is their primary the structure or their behavior ontology is going to be and what kind of visual patterns will best suit for that entity circle this is the visual patterns we prefer to show to the user and some examples tools and why it works, why that pattern and all. With this information what we have tried to understand is like we try to map the data type because we broadly at a high level it's like a qualitative and quantitative but even in the qualitative and quantity we have multiple data types discrete fraction ratio and something to store the in the qualitative you have spatial and uncertainity. 04:08:50 Manisha Gundapuneedi: Now each data type has its own way to represent to convey the meaning and intent and here also we said what is the right way to show this thing and now we try to map this to the archetype that we have done we have previously established. So so far we arrived in the clarity in terms of persona and what data will come and if the data comes what is the visual pattern that need to be shown to the user. uh and per this is another thing with that context what we have done is like we here not target to the personal but what we have done is like if someone is coming to the platform for the first time and they fall into certain archetype doesn't matter a reputation maker or a deliberate one what kind of information can be exposed to the user it can be a daily briefing someone interested or they want to understand what is a drift uh is going to frictions that are going to create on the system or what are the blind spots so these are different areas to what can be exposed to the user as a beginning to start the conversation or to initiate the things. 04:09:49 Manisha Gundapuneedi: It typically based on the persona archetype system will pick any of those things or based on his pattern system will automatically bring these things to the table. Now we are here. From here what would be the starting point? Fantastic. That's where we are. We need support. Fantastic. This is what is needed. Okay. From design we feed these guidelines to the AI system. Okay. Let's say the first very first step for example is because you don't know anything about any user and all of that the default is daily briefing. Yes. Now daily briefing what matters today. The system will figure out what matters today and the system has picked out some data elements. Based those data elements it will follow the guidelines to say how it should be represented. It will make the decision. There will be uh for this you know this is the widget this is the widget this is the so sorry so is presented uh the work that we have done right so I I'm saying we feed this to the LLM these are guidelines okay if a person is this arch type this is the kind of behavior they see this is the kind of and if they are asking these types of questions they most likely are this persona so you you categorize. 04:11:18 Manisha Gundapuneedi: So so first of all based on the questions they ask based on their interaction patterns system will over time figure out what their true arch type is. Then you have also said if the data is this way then the best representation is this. If the data is this way the best representation is this all of that the system understands that. So then you said hey if the you know the context is daily briefing this is what you need to do this is what you need to do like that so these are the patterns that these are the the use cases that I'm asking so the the system will decide that oh because there is you know navin is this arch type laidback you know lazy fellow okay so this for this arch Right. Um uh you know what they typically look for is this and uh because this is the fourth time he's coming to the platform the daily briefing has been done. So now in this particular time context what we need to to show him is this. Let's say which for are we sending time correctly is the question that he needs to answer right now generally. 04:12:34 Manisha Gundapuneedi: So therefore this is the kind of from whatever my knowledge graph is this is what I want to show him and because I want to what I want to show him is now decided it will then say oh you already fed me the guidelines of if the data is this then I have to show this if the data is this I have to show this now I am going to make a decision based on everything I know here is how I will orchestrate the UI you know I will show uh these two widgets side by Right. And this widget below them and this CTA here. And how that layouting also is done is also based on some guidelines in my mind. Okay. It will consume that and it will throw that to the UI and say bus this is what you need to do. You know this is the order of widgets order and the data sources are this connect them up and show and the UI shows package. Actually this is what I was thinking. one week or five days or something. 04:13:40 Manisha Gundapuneedi: I'm thinking this may not be enough. We may I mean in order to while I while I am with you, we need a proper system in the background. Okay, this is how I was thinking Rakkesh and I have not solidified anything yet. There are certain like for example Gallup what is all Gallup about? BP10 is about a CXO level decision maker. There there is some psychology and philosophy behind why they came up with 34 is all about strengths. It is all about how a human thinks right. So I'm thinking maybe we need to use those kind of existing proven uh frameworks to define archetypes for enterprise brain. This is how I'm thinking because it's proven or there there is some amount of understanding as to how a person is right and then I'm just thinking out loud. Bear with me. This is one way one way I was looking at it. The other way I was looking at it is um strength strengths analysis. There could be another frameworks for defining the how a person galloper is all about thinking Rakkesh. 04:15:16 Manisha Gundapuneedi: Okay. End of the day, Gallup is all about thinking and murder. So when when we when somebody presents let's say a statement to you, uh whoever has the restorative on the top tries to immediately solve it. Somebody who has uh uh uh uh deliberative on the top tries to find the flaws in it. Somebody who has uh oo on top tries to influence the person and say everything is going to be okay types and you're understanding some so based on the strengths that you have and first they play interchangeably you use one of this. So is all about thinking they might be feeling frameworks, emotions. What triggers what emot how they take decisions. Okay, this is one part. The other one I was trying to look at is flourish. They're more of still information uh patterns. Those patterns are still giving you information. They are not decision uh uh uh driving patterns and how to present information. So I'm wondering if there are any decision driving patterns out there uh and what would that be? 04:17:20 Manisha Gundapuneedi: So um has lot of decision making frameworks which drives everyone to start taking decisions. So this is the research I thought we should do. I I wanted to discuss all this with you yesterday. I didn't get time to discuss with you. So I I I don't know. I'm just throwing more variables in our direction. Thank you. This is friends and advice. It doesn't matter. You have any anything to add this? I already responded. No, I don't think you you you have any opinion on whatever I mention. I already and you're saying the that is enough and whatever I'm saying is not required. No, I didn't say I don't know that you even heard what I said because you didn't there was no acknowledgement response. So I don't know whether I should what I said whether it made sense or made sense. I don't know. I was not part of whatever you said. Right. Even when you came back, I explicitly acknowledge and said, "This is what we're talking about." 04:20:53 Manisha Gundapuneedi: Sorry. mostly or maybe not. I didn't acknowledge. I don't know why I would not have acknled. So you want to repeat once? You want me to? I'm saying it's great that you came up with these patterns. These are which patterns? Whatever you show. Okay. Okay. Basically, this is the framework that I believe we should feed the LL that hey, you know, people fall into these arch types. This arch type behaves like this, you know, this arch type behaves like this, this arch type will ask questions like this, things like that. Uh, you know, that is one side of it. The other side is generally this arch type person will look for information like this. They they respond to information like this. All of those that you have already mapped. So that becomes the multiple you know you have multiple each of that has valuable information that link together will the LLM can make a decision. So that these are guidelines for the uh the system uh to make those connections and say oh because this person is coming and this person generally is this arch type um because it knows that because it has watched that person for a while so it knows that arch type and it knows you know this at the middle management level so they care about this kind of stuff. 04:22:41 Manisha Gundapuneedi: So because of that entire framework, it can make a decision on um sure and correct. So it it can make a decision on um at that point in time what data needs to be you know based on that arch type also based on what is important to them what information is important for them to uh for what information needs to be presented to them and what kind of uh decision options need to be presented etc. Everything that needs to be presented it knows and therefore based on the framework that you have given it can determine that oh for this I should probably uh do a bar chart for this I should probably do this and so on. I'm saying that we a week back we both agreed that that is the approach but now I'm saying we need to change the approach. Okay. Okay. I'm not debating the architect. I'm saying how we define the architect should be different. So standard framework this is how these people think this is how they feel. This is how and if you are able to connect the dots then based on that archetypes and the personality that is one thing. 04:24:15 Manisha Gundapuneedi: Second thing um can I respond on the first I'm saying and that is fine that doesn't change what the I mean that is just you're giving different inputs to the system I'm fine you come up with whatever framework right because but I like the framework itself that you're defining these are arch types these are their behavioral patterns this is what they care about this is what is important this is and mapping of this to this what what kind of stuff is important to them what questions will they typically ask that framework whether you change the inputs to that framework you add new dimensions all of that is great that framework will help the LLM put together the oh what should be the UI for LLM to put actually I gave a different framework did you discuss the information archetypes I showed everything information archetypes okay I'm saying all together because one part is influence. Did you show the information of these things and not Influencers influencers visionaries influencers archetype. How to define how the mobility the depth of data The strength of data data influences influencers and bitcoin type of data data influenced. 04:27:06 Rakkesh Yenugudhati: and the date of the Malik. Manisha Gundapuneedi: output oriented. Yes. Okay. Mhm. archetypes persona archetypes. First defining the entire user base into 10 archetypes and uh defining that what on what basis we are defining into particular archetype and what are their code patterns and then what the decision style and what they prefer in terms of decision making and we will leave the other things each archetype we we identified that what are the different personas that are going to fall into even though they're rapid deciders we are still working on points I thought it's done. Yeah. So here we made that what is the dominant role even the archetype person but the approach is right we'll just revisit it and first of all what are the different attribute that defines an architect do not put a personal are all influencers attributes of an arch archetype. So at the beginning it will be very few but later it can be matured because I I just put few but we can add more. This is fine. 04:29:05 Manisha Gundapuneedi: And here is the information archetype where we are telling that information itself we put it into an uh 10 different archetypes telling that it's an entitycentric that means event entitycentric evententric statecentric time and distributional rational spatial text semantics like that we define the data into 10 archetypes and we also define that what will be the structure of the data and what are the possible ways to represent that particular data in visual format. Right. And why this work too? Now after this what we have done is similarly how we did the personas there. We try to understand the data here. So we data in terms of qualitative data and the quantitative data in qualitative itself what are the different data units will come into the picture and which archetype it belongs to and in order to represent this particular data what is the supported visualizations need to be done. So after that we because considering the all the thing what we have done is like what would be the initial landing page would be considering the archetype what what are the different experience that we can give either we can begin with the daily briefing or we can talk about the what are the blind spot find post the login last login to now like that few states we have defined When the user login for the first time based on the archetype and the persona what information will be exiled them 04:30:39 Manisha Gundapuneedi: because day briefing is not one time right because you can't brief every time. So we need to find when he's logging again to the system what are the information that we can show This is an evolution of this right now. Mapping of information to the person archetype. One person can be associated with many information. I know. Okay. Yes. Yeah. So information right? Yes, that influencers who going to decide this multiple frequency of collection of data is an influencer. volume data mapping layer below what output should be output. information. Who is going to use the application? We start that who is going to use the data frequency of the data is connected to that person. Now set Important patterns. data patterns. data patterns already. So dat what you pattern can be changed in more into viewing the debriefing or interested to know the more debriefing already is an output pattern. 04:35:11 Manisha Gundapuneedi: Look at day lifting as an output pattern. Okay. Uh how you show man? Okay. The debriefing is what you show. Uh day briefing is the output and what you show how you show. Usually the output pattern covers the covers uh the five W's and H right. Okay. Combination of five W's and Hutn existing patterns of visualizations influencing factors. right scientifically already proven. Nobody will question us on ours. That's where I'm coming from. because they're already proven and we are relying on them to make our conceptually this is what we want. don't observe all of this uh yet. Sorry. I'm saying I don't I mean right this moment I don't need to I mean right this moment I'm not able to but we do need to absorb it to actually implement. I'm requesting you don't do it yet. Huh? But I am trying to make a different point. I so uh but this is the framework that we need to be able to develop a a first but while they are working on it. 04:37:45 Manisha Gundapuneedi: Okay. But what is defining the attributes? Okay. He's picking up your rapid. I'm saying the framework is good. What framework they're fine with that? What do you come up with whatever information architect information and data I am saying you have concept I feel like you have come up with a framework that is able to connect the dots okay if there is a dot missing if the line missing you know it will come it will become apparent when we try to really absorb it and then we'll highlight it you will fill the dot I'm not worried about the dots changing spaces where basically from here to here you have multiple frameworks that connect the dots. Okay, you can change them however you want. But in the end, you will give us a framework to be able to decide at this moment you know there are certain number of inputs about a person, about the day, about the time, about the context, about the information. There are certain number of inputs we pump into this framework and then output will come. 04:39:23 Manisha Gundapuneedi: The output will decide what UI should be look like. Right? So you can change whatever you want in it. There is inputs and then out will come the outputs. That is what I see from this framework. My while you work on refining this framework. This is what I want from Manisha and Rajar. You please work with Gopal Garu and Yashwant. what I want at this time is a simulator. Okay, Rajashekar G: Okay. Manisha Gundapuneedi: this simulator in my mind can be built uh simply based on uh you know you can do charge GPT also you don't even I in my mind you Rajashekar G: Okay. Manisha Gundapuneedi: don't even need to code it but if you're if you're writing a small simulator connected to OpenAI or Germany fine please go do that and there it will basically absorb this framework and you will uh your input to that simulator Rajashekar G: Okay. Manisha Gundapuneedi: is Hey, here is the situation. You know, it's about 400 p.m. in the afternoon. The person has the the uh Naven is uh Naven is the head of delivery and he has come to uh you know this is the third time he is logging into the system. 04:40:41 Manisha Gundapuneedi: Okay. Um and uh you give them the uh since the last time he has come in here is what happened in the system also is an input because that will come from the data layer uh you know the intelligence knowledge graph layer. So you are also inputting that then the simulator based on this whole framework is generating here is the UI not the UI itself it's just describing the UI. Oh, the first uh you know the two widgets on top will do this then the 10th this one then the CTA it will present three CTAs to the user etc etc all of that is actually described in the framework right because what is important then that is broken down into multiple data that you need to show and then the data is broken down into what is the representation that is right and so on so I just want to be able to prove that we're able to process this frame framework and based on this framework dynamically generate a UI right so the the you know Rajashekar G: Okay. Manisha Gundapuneedi: Pratima and Rakkesh will refine the framework so we'll just change the rules right you know instead of processing you know uh 30 rules in four spread you know four sheets you will process 60 rules in six sheets but it is just feeding in prongs to the lln And so therefore suddenly the behavior will change the UI behavior will change. 04:42:18 Manisha Gundapuneedi: I'm saying for the to simulate the data also I will give it a context. I will give the data as a context. Oh we have a new lead. We we have this. Oh there was an escalation. Blah blah blah. So I will describe some level of data um some level of context to the LLM because this first of all this is a simulator right then it has to so I am looking at this simulator to be done in 2 days not like 2 3 weeks okay then we are able to prove that hey we we are able to understand the framework the framework can change but the system is able to process the framework it is able to interpret the rules in a particular fashion one by one by one. So given a set of inputs out comes a UI and it should make sense to you why did the system take such a decision it should be clear to you why it took that decision. If you're able to simulate let's say 100 scenarios like that and then you are seeing that okay in the 100 scenarios the UI the the UI that it is describing is what we would have chosen in that scenario then the system is the simulator is doing a good job right the system is doing a good job now basically that is encoded into the actual uh orchestrator UI orchestrator right and then then you all you need to do is connect The simulator how are you going to build? 04:43:49 Manisha Gundapuneedi: You need to tell the simulator all the framework has to be there for simulator has to input framework. Correct. Processing. See in my head there's an input framework processing framework and an output framework and all of these are driven by multiple inputs and agents. This is how I'm looking at it. Maybe you can explain what are input framework or processing framework. Input framework is people. People people um um people is an input. U so far what he has done or has been doing which is the process or information workflow is an input. Um what are the external stimuli are also an input? Inputsuming is a simulator is full of agents. Imagine them as those what are agents in your mind. Think of them as minions. Okay. What do they do in your mind? They will pick based on the inputs that they have been given. It will pick up something like a Lego block and he'll bring all the Lego blocks together and different agents will bring all the Lego blocks together and they'll decide what to build and it it'll give information and I'm still talking information not output just to be because we don't get confused by semantic because we've been using agents in a very very very different context. 04:45:39 Manisha Gundapuneedi: So maybe pick up another let's pick up another thread maybe uh so let us say uh let let's say they are not because they to me they are just they are not so much agents they are just rule interpreters right is that what you're talking about let me put itern what I'm trying to say is last time layer of room full of holes in water just trying to do this. See end of the day it's all about prompts right the better prompt you get the better output you get so prompts or the simulator is full of prompts simulator is full for example you talking about these frameworks so just to step back This framework is actually saying, hey, typically an executive who is a uh give me an architect rapid decision rapid decision maker arch type plus a CXO uh will need this information or or this kind. So you're defining conclusions first draw something you're defining. mapping. Okay. Delhi to mapping himself for him to get clarity or for me to get clarity a strict mapping is okay but it should not be so strict then the whole point is lost okay I I am okay with that okay I'm okay with that not having a strict map so input is one input to the simulator. 04:48:23 Manisha Gundapuneedi: Yes. Then uh the person archetype and person is one input to the simulator. Correct. Then the what uh so far his his patterns behavioral patterns is one input. Yes. Uh his preferences is one input. Yes. His thinking is one input. Yes. on one one second though he's thinking I am able to collect and record because his behavior pattern based on his interactions the kind of questions that he has been asking me the kind of interactions he's been doing on the platform I'm able to record he's thinking I'm not able to record he's thinking I won't know unless he expresses Ah are they a yesu sir he you can based on the type of questions he's asking is thinking me that is why I differentiated the two the type of questions if you're saying oh thinking is basically type of questions I have that it doesn't have to be strict that is the whole point gopal also was making it can be very fuzzy I don't have to strictly say if you ask this question that means it is this kind of thinking if you ask this kind of question that means means it is this kind of you don't need exact definitions like that we do need oh generally these are the questions that somebody ask if they're asking these kind of questions generally they are interested in this kind of information generally if they are interested in this kind of information this would be the kind 04:50:12 Manisha Gundapuneedi: of uh uh this would be the right time to show that like those general guidelines is what the system LLM needs as skills to be learned. You can always keep changing the skills you so you can always keep changing the framework. Okay. So all of these things in the framework let us call them skills. Okay. You are the system is learning these skills. It is learning that first of all there are 10 types of arts types. you decide you know three months later that actually when we go into production we have 11th new architect you can add it. Okay. Similarly, oh this is the um you know hey if you have numerical data lot of time series based numerical data the best display model is this that is a skill. So you the system has learned that and you have fed that similarly whatever the intermediate things I've seen I've seen like four five sheets of in the end what I'm looking for is a loose connectivity of and it doesn't have to be strict onetoone mapping very fuzzy connectivity of generalizations because the LLM has the ability to synthesize those generalizations and based on the context Next, come up with something. 04:51:40 Manisha Gundapuneedi: So, I'm not looking for a rules engine where you say strictly at 7 a.m. when the CXO comes in, you need to show them this is not what we are looking for. But we are looking for a rapid decision maker. Generally, you know, the first thing they want to see is something, but if they're repeat visits, they're looking for something else. So that kind of general guidelines whatever framework you deciding that framework is what we need. I I I think we need more. Okay. That's what I have seen. You need a thinking layer. You need a feeling empathetic layer. You need intelligence layer. You how you are going to get it into the simulator is something you need to look into closer and closer. Now that is my belief. Okay. As a human being, how you would make that that intermediate layers, how you would make that? Are you able to explain as a human being? Forget codifying it. As a human being, can you make those things? 04:53:16 Manisha Gundapuneedi: I'm okay. If you are designing it for a particular context, how do you make that decision that whatever you said there is an emotional layer, there is a thinking layer. What do those layers mean to you as a human being, as a human designer? Are you implicitly running that in your head and coming up with something that intelligence layer that you're saying you have to build that that is what we want to build. But what is that because I need to know I need to understand what that is. What is that intelligence layer first doing? Did that make sense? I was I don't I don't know if you made sense to me. So that is why decision making frameworks we need to go back to the decision making frameworks and what are all the cog wheels of the decision making simulator. It is one of the frameworks that the simulator has to use. They will be thinking there were lot of mad along with it uh decision information how to present an information what are the decisions they will take in the context of it and what could be the future drill down of information is how I I am I am thinking what lack of a better word from nama co I think that is that is where I'm coming from and I'm building that simulator 04:55:08 Manisha Gundapuneedi: I want to build that simulator okay what I need is whatever you're saying hey you need these decision making frameworks you need to understand the subcog wheels you need to then decide what general guidelines framework I need he said a few things I'm sure because yes is not there I can't comment on his behalf if you if you're um okay I I Maybe he has knowledge on that and that is what he has to but what I'm eventually saying is that is not a coding thing that a techn piece of technology will do that is part of what you No it is not part of technology please please listen to me it is not part of architecture or technology it is part of what you have to feed into the system as a decision making framework work. Okay. So, it is something a human has to sit down and define and then at the at runtime the system will interpret all of that, take the context into consideration and then come out with the actual outcome. I mean they need that people can't do that kind of a research work or guidelines people can't do research or guidelines I mean technical team is not able to understand the personalities what guidelines what rules and visualizations maybe what I'm saying is maybe yes is able to give that Rajika is capable maybe Manisha is capable what I'm saying is in the end it still needs to be codified into guidelines it's not something that sit and code in 04:57:38 Manisha Gundapuneedi: Python that is what I'm trying to distinguish between it is not the analysis analysis as as intelligent people if you want us to contribute we can contribute but it is not a piece of technology that you can build that will build that intelligence layer is the key aspect I'm saying do you understand now if not I'll explain it is knowledge knowledge documents think of it that way it is basically a piece of document that you have to give the system to say hey this is how you should behave and it'll behave okay it is not something see AI is just it is just a prob probability machine. It has no intelligence. Okay. AI has no absolutely no intelligence. It's a probability mathematical probability machine. You have to feed it what is the the the the thing that it has. You can of course you can say hey we have defined uh these 10 archetypes. Okay. Because LLM has all of this anything that you talk about frameworks all of this knows there's nothing that you know if you have read about it it already knows okay you can say tell me how these archetypes behave or the behavioral patterns all of you what of those are applicable for our UI basis we have to somewhat reinance around that you shouldn't you can also leave it completely to the system but then you'll get very arbitrary 04:59:21 Manisha Gundapuneedi: resp you can use charge to do all of your research I'm not saying no and because it is charglm when you later on when you feed also it is very easy for it to understand you don't have to explain each and every arch type you don't have because it already it is the one who gave to you so it already knows You're just saying some guards and you're connecting the dots. You are saying because this determines this and generally somebody who behaves in this way will will ask these kind of questions and if this is the context then this is what you need to do and because if this is what you need to do this is how you expose that information. So you need to sort of connect the dots for the LLM and therefore all of the connecting those dots is going to be done in you know for lack of a better word prompt files right so you're teaching them the skills then it is going to learn all of this framework skills not that it didn't know it knew the raw information but you have supplied it with the connections how to use this piece of information this piece of information you've connected the dot And see that is see if you ask lovable hey I'm building a customer service system please uh you know generate why is it able to generate 80% good UI because it is already encoding all of that so LLM already knows a lot of this but you are setting up 05:00:58 Manisha Gundapuneedi: the the framework the way that we wanted we understand enterprise way so we need to do In the end it is not something Raja Manisha or Yeshuant have to code either we prompt or the LLM already knows. Okay. So there is that is the advantage that we have with LLM. Most of it is not code most of the coding that we're doing is basically connecting things you know absorbing information indexing it storing it. So the use in the end the heavy lifting is actually done by the LM feed all this information it is actually generating the response that is why the response is so arbitrary. Okay. So while the raw information may be known to the LLM, you have to connect the dots so that the guard rails are set up. That is not something that I would want you know Raj or Manisha to do from an engineering perspective. If they as individuals if they want to think about that I'm okay but there I'm saying that we should do the if if need to contribute to that workshop as well. 05:02:29 Manisha Gundapuneedi: Is there something that you you want further? you you have questions or you have no confusion system and system architecture. Maybe you're using the term system architecture in a very different business architecture or I I don't know if you're using that. I don't know how what in your mind is that but architecture for me is something different. So I can't address that. So frequency output information should be challenged. Thank you. Okay. You are saying okay there are 10 you will you you might change it. You might even ch change the definitions or the patterns for each archetype, the decision style you're saying. All of those things could also change. But there is a framework that you will work on. In fact, you ask the LLM, it will give you some framework. Is there a question here that there is a framework that you you're okay to give this framework? One main thing will deliver the archetypes. This is basically the end user archetypes. 05:05:27 Manisha Gundapuneedi: Okay. What is next? Further drill down and all the architect to multiple. What in in your mind? What is this defining? Huh? Subcategories of a given archetype. These are subcategories of a given archype. Why is this needed? Because we we are thinking that user will be the the rapid decision maker will always take the quick decisions quick but some people are really interested to know by and some people directly believe the system and they can able to take a decision they want not only the insight they want an evidence little evidence to that person we put in a subcategory of an archite Yes. Now um and why is it needed in this context that you are you saying that oh because of a rapid decider uh for a pattern commander the the type of information you show hard skills and soft skills the architect is defining the hard skill based you need a subdivision for soft skills so that you break it down rapid decision maker and they are quick to take decisions but what they need to take a decision changes correct so I understand how the first one the first tab you showed and the second tab you showed I understand the connection yeah okay next one I see thank you we are the role they're coming into which is a dominant personality that so now you are train in a corporate because you kind of know whether somebody is an executive versus a manager/ 05:07:17 Manisha Gundapuneedi: operational manager, supervisor versus an individual. For example, decision making decision basically it is all about certaintity of information. The more certain information you have, the faster assuming that they're not legards and memories that's a soft part of it. So framework is right. This is the first cut actually he made right again the the please don't uh yeah yeah yeah and I am okay this completely changing the whole frame of changing also I'm okay because in the end sorry in the end the we have to teach the system that these are not absolute rules these are all probabilistic and the system will run probabilities The whole LLM is based on probabilities only. So that all it will do next and I understand how so I mean if I take it as after define and it is not connected because every tab is disconnected in some sense too. Okay that now we're coming to something like that. So can you just go back? So for example, I mean if I just take if this is the right framework and you stick to all of this, I would interpret this as hey you know when the person comes in um you know if they are managers and look at their dominant you know they're either loss avoidant guardian or a deliberate analyst. 05:09:03 Manisha Gundapuneedi: So therefore do some I mean they're typically that is the let's say context I'm giving. So I understand the connectivity to this. Next, how to define that? What is this attributes of a person? Because if you said right, I will put this person into one particular archetype or persona under what? So these are different parameters or attributes that are considering because they're very action first but when you when you start interacting with the pattern with the system and all let us say I absolutely know I'm just saying let us issue that the system knows the based on the past behavior of the user uh the system knows what is the type of archetypes type sub type all of that let us say the system knows okay um and so what are we saying here here what we have done assume that for the first time I didn't even interact with your enterprise bin I'm coming to the system for the first time so when I come to system where you want to put me in I know you are already decided Look at this. 05:10:41 Manisha Gundapuneedi: Restorative antio. Okay. to harmony then may not be quick decision makers as as an example or uh arrangers uh may be a quick decision maker or something on those lines a mapping chair archetype to personality mapping chair this is an attempt to make the personality mapping personality mapping first you will need it over a period of time because they are the influences will define which decision to throw at them. I understand. Okay. No. And because I need to build that simulator. I want to build that simulator. See goal personality challenge there are attributes. Attribute based personality. personally person. So as an example, so are you suggesting that? Okay. Because a very directive query style typically is a rapid decider. Restorative is one restorative is one attribute. So problem that guy's personality is so he comes into one of the one of the attributes that contribute to a rapid decision maker. Rapid decision maker is restorative. Correct. How do I know? 05:12:43 Manisha Gundapuneedi: Okay. Sorry, I thought you were done. That is why I was responding. Okay. Go ahead. Finish. Then I'll respond. So it is trying to show him a solution as a decision because what are the rows and what are the columns? Fine. Restorative LLM understands what restorative is because it understood that behavior. Okay. Is there a guideline on how I identify that somebody is restorative? Contextual documents. If you can map it, great. If you can't map it then you will take by default. This is the important thing that agents whatever you want to call they will decide this personality is restorative. Let let me again you bring it back into the context of um an enterprise brain working with an enterprise all of that and you're building something generic uh some even if you give access to the you know person's HRMS system right um 90% of the time you will not be able to get whether the person is restored the strength of uh so that that is what I'm trying to determine that is what I'm trying to showcase how this thing is happening don't and if something is generally not there at all the framework it will have very little value in the framework because we will not over a period of time depending upon how he's working with that is 05:15:05 Manisha Gundapuneedi: what I'm trying to get so am I a that is not what I got you are saying restorative is an input if you're saying that restorative I have to figure out that I'm okay with so you so this framework will dep will determine what kind of behaviors typically will indicate somebody's respirator or not are you saying that then I'm good thinking huh thinking how they think is what it will come sure thinking age of After that. Okay. What is his feeling? What is his thinking framework? How he decides in a framework? How how what are the type of actions he takes in a framework? Then that is what is his thinking framework either I know that this person has this thinking framework or based on his interactions and his behavior it over okay I'm okay with that uh how is his what is his action framework that is an output right because what drives his actions and framework what drives his action what is the what is one dimension what is other dimension give me that just just for the you can change it give me an example and then what action framework hey restorative person will take these kinds of actions on nava or are you saying this point okay whatever it is you will define a framework Yes. 05:17:21 Manisha Gundapuneedi: Yes. I'm saying one framework is about his actions framework. One framework is about trying to figure out his thinking. Right? that right how he trying to figure out how he's going to feel trying to figure out how he's thinking trying to figure out how he's going to react and decision framework of I don't want to frustrate you I am saying there are two distinct things one is about categorizing a You are not accepting that even that action thing you are in your head feeling like action thing is driving the output it is still driving how he's going to act the the decision making framework is the final thing that will give the output but that doesn't necessarily still decide how the user will act because user end of the day acts his own way correct okay you are We will that okay how he is going to think how he's going to feel how he might act the probability of acting based on that what are the what type of decisions he may take will drive it how he is active reaction. 05:19:14 Manisha Gundapuneedi: You're thinking in some form, you're feeling some form, you're reacting in some form. probability of that reaction, probability of his thinking, probability of his feeling are the these three things until is is I am I able to measure his reaction in the system or not. That's the only thing I want to ask. Not yet. I'm not able to measure his reaction. Right. There is no way for the system to understand. Are you are you able to measure his thinking? Huh? then you'll be able to measure I'm trying to tell you there are three frameworks okay now what words I want to use I don't know because a word you you okay three frameworks three frameworks which will drive how a human what a human may take a decision the decision is influenced by three frameworks three frameworks. Now for each of those frameworks are there specific categorization. First framework five six types second framework 18 types third and there are there are overlapping distinct labels. Yes. Now I am able to identify a person belongs to in the first framework he belongs to this category in the second framework he belongs to this category in the third framework he belongs to that category unary over time I'm building based on his actual actions on the 05:21:02 Manisha Gundapuneedi: system they're not independent they are very very dependent could be dependent also could be overlapping could be mutually exclusive could be independent whatever it is I'm not defining a rule but we are Okay, we are able to the system should be able to figure it out based on the based on the actions and the type of questions he ask and based on his behavior pattern on the system. Yes. Right. So now we do need a fuzzy mapping of if the person asks these kinds of question mostly they probably belong to this category in this first framework. I am saying that is needed. Okay. If because if you are saying in the first decision in the first framework there are six categories. I need to understand how to identify a particular person and categorize them in those six categories. You can say that pick a default and build better. But to build better I need to understand what will make them into first category versus second category versus third category. They don't have to be very very very strict rules. 05:22:24 Manisha Gundapuneedi: It is fuzzy. Mhm. That is I'm assuming this is that next what so now I'm able to assign them categorization across the three destin frameworks three frameworks. Now what what is next? Information architecture. Now we've gone into information already. Let so is there a connectivity between now you have three uh frameworks into which I have categorization. Now roughly people in this category uh uh people of type uh uh this category in this decision framework uh will generally ask for this uh people of this can ask for this. This don't have to be mutually exclusive. They don't have to be strict rules. Again they can be overlapping. Will there be a mapping like that? right now. Okay, I'm okay assuming that all of those are there. Okay, so I want to prove that this can be done. The UI can be built like that. Go ahead. So while you work on refining those frameworks, I will take this framework and build a simulator. 05:23:52 Manisha Gundapuneedi: the right framework. It doesn't matter. So should we just wait until you're done with your framework? While you're saying that it is very dependent on these frameworks you're also saying I'll go ahead and build with this that is what is confusing already thinking framework thinking frameworking framework and reaction framework you want to pick some examples pick examples within that why are you saying the because what we don't know we can't you know imagine Okay, please. If you are willing to you know if you have a back and forth discussion I'm able to explain to you how the simulator will be built why it is you know one pattern will work why one pattern will not work or you know what will help what will not help or why I am saying oh it doesn't matter even if you change the framework why I'm saying it you will Understand if you're not willing to engage and you're getting frustrated that you're getting frustrated then I'm not able to move no no this framework is wrong what should I do if you want to take it and build it go ahead So whatever framework you're coming up with whenever you're ready then let us know then we'll build it. 05:26:10 Manisha Gundapuneedi: Alopa maybe you can look at okay you know there are those visualization patterns visualization patterns key information archetypes information is a question data types So that we look at those output types right based on what is a good library that will support most of them what looks really good really fancy link that nothing should be from okay this is where I'm seeing that you will again get a static static dashboards the current dashboards you will get a a little better experiential information dashboards they won't be decisionentric dashboards you guys are following whatever I'm saying dashboard I'm calling them information based dashboards multiple graphs list views tables information that's the name I'm giving it okay flourish patterns they will still become information dashboards but experiential information dashboard dashboard information dashboards right now they're leaving at certain place it's not allowing me to take a next step of actions dashboards because everything is like we know that kind of a person. So for example actions that he can drill down informations. informations plus wisdom. It's not decision but every so That's right. 05:29:51 Manisha Gundapuneedi: Hear me out properly. Flourish patterns. What is still getting highlighted through flourish patterns? Information. Okay. I am challenging ourselves. So, think of those dashboards as information first dashboards. Now by putting flourish patterns they'll become experiential information dashboards or experiential information first dashboards. I'm challenging ourselves. I'm challenging ourselves can we make the dashboard decision first dashboards or decisionentric dashboards where information becomes secondary decision becomes primary is my take. Now how it is going to be what it is going to be for examping information but they should be able to first see these are the five decisions if they take their business impact will be like 10x something along those lines that is what I'm saying list of list of decisions these are the five decisions you have to But I feel like how do I make a combination of decision, insight and information in such a way that decision gets highlighted, supported by information. Information gets highlighted and then we say look at the information. 05:31:54 Manisha Gundapuneedi: This is the insight. This is the conclusion. For example down they will do is secondary rank. But end of the day. All of us in our own level the decision. what drives each person's decision but end of the day decisioncentric dashboard and it should empower and enable people to take faster decisions which help them save time end of the day everyday at 4 every day around 4:00 or 6:00 she used to order some junk right so and then there used to be a patterns with her if you realize sushi once a month because it used to be expensive so there used to be certain patterns so you would log in at 400 p.m. on a Tuesday, right? literally pani from this one just if I change my mind why every Tuesday in the last in the last 10 Tuesdays you have only ordered pani puri and henceforth I'm suggesting this to you however the alternative order is this if you want to change your decision that is decision driven uh Um yeah one thought one decision basically like for example let's take the sales example what you're saying is our dashboard should be decision centric dashboard currently uh it says that you know um uh sales is not performing well in South America that's information that's information so decisions and dashboard you'll say that you know there are three decisions that you can take on this particular thing and if you for the first decision this is how it is going 05:35:09 Manisha Gundapuneedi: to fire the saleserson in South Africa is a decision that it'll tell me then it will say some are descriptive some are you can also show it as a visualization see visualization is a mode and then think of it as raw content you put content on Uh description is this this raw content audio encrypted something you're understanding those are the uh the channel or the type through which you will express graph is just one form of expression. Table is in one form of expression. Yeah. So graph is just a form of expression of the information. It doesn't necessarily have to be a graphic. It can be an audio file or it can be some snippets of images. It can be a video. It can be anything. Don't imagine it only as graphs and all of it. It can have a video. It can have an audio. It can have anything in it. It can have an infographic. It can have a uh a visual graphic. 05:36:47 Manisha Gundapuneedi: It can have a illustrative graphic. It can have anything. It can be anything. I want you guys to imagine beyond the regular dashboards. It will depend upon my see for me right anything I like to take notes because I'm a kesthetic person right that is me is an audio person I have to give an audio feed to People write because they want to feel everything. Audio people or auditory people have to listen. Visual people have to see human don't look at it as only graphs or just go ahead. Go ahead. So in the end there are atomic uh whatever you want to convey whether it is information or decision or uh insight whatever it is there is finally an atomic it could be visualization it could be audio file it could be what there's an atomic what are the atomics right now uh I I want to be able to build the UI with the whole bunch of atomics. The orchestrations of the orchestration of the atomics will come from the system. 05:39:03 Manisha Gundapuneedi: No you say instead of atomics you call it a Lego block. So the system what do you want to call it the fin there is some work that we can do in the meantime in the meantime we figure out just first of all I don't want you to imagine a dashboard to be just a bunch of graphs and tables and tables or uh uh you know alerts or things like that. That is fundamentally something I want us to move away from. So depending upon what kind of a person he he listens to it and his absorption if he wants to change then probably you need you need a motion with visuals. So first of all outputs I need you to expand beyond graphs that is one point. Okay this is one output. What are other atomic units that you want to describe? CTS, data collection items, um data collection elements, um information, audio, video, everything is anatomic. Okay. We all tend to we all like to categorize stuff. I'm trying to give a direction because I'm saying okay in the end what is everything that I heard so far is sounding like okay you have to have an extremely flexible UI okay there is no structure to it there is no definable because right now they don't even know whether they can define a structure not structure whatever whatever whatever right so it is almost like hey the UI is made up of 2,000 React components that 05:41:34 Manisha Gundapuneedi: can be put together in any combination put together this combination that combination right now that's all you can do so just as again I come back to the simulator right first things first I want the simulator to be able to simulate a uh you know if you if you define the framework Whether correctly, incorrectly doesn't matter. Based on the framework, can you generate a URL? What framework do we have? We have whatever framework we have today. Okay? Which we don't understand fully. They are not willing to explain it to us fully. But doesn't mean that you get to distract. You don't get disturbed. No, but you are talking in the middle. We're not talking. You are. You're saying you do not ask. So um so this is basically a collection of you know uh things that you are able to arrange right. So now I want a simulator. I want the simulator to be able to take in a framework and based on the framework and a set of inputs able to simulate a UI. 05:43:15 Manisha Gundapuneedi: We don't have to render the UI. Even the description of the UI or the order of widgets, the order of these things will tell us that oh this kind of thing is possible, right? The LLM is able to absorb this and then give a decent output, right? So we need to current framework whatever it is um when are you thinking that you'll be able to finish the framework and then just give any based on that we'll make a decision next week. Next week Monday morning yeah that's fine. So that means that we have that right. So there is some framework right now. Okay. Some parts of that we may understand we may not matter just simulate that framework also. So that we if we change the framework the output will change right. So I want the simulator to be built to be able to prove that we can come up with a dynamic UI. Then all you're doing is on the UI side you have to be able to display a bunch of atomics right now uh that atomics display I mean the atomics definition language do we come up with our own is it like very complex JSON we want to make that simple right we don't want to over complicate that come up with something like HTML again yeah okay that is too much So for that we can wait for a little bit more for them to crystallize that 05:44:54 Manisha Gundapuneedi: but we can work on hey what are the right atomics? Are we able to display the right atomics right? There is um uh one of them is okay you know but the best visualizations where are we getting that from link that flourish you know anything else like all the 3D kind of uh uh components and all of that you know link that as well so that we are able to but then you need to at least build a a basic framework so that if something like that comes you are able to display the bunch of atomics right so I don't know that we can do it completely without fully understanding the different patterns possible but at least we can make a start that is what we need to do right so that we are at least some because this is not trivial that UI that building that very very dynamic UI is not trivial that will take a little bit of time so do you think so what is this fogg versus this man so we concluded that we'll like react would it be capable of uh dynamically be able to put different components in and on the canvas and then it is capable of that are we required or we are done it's up to meeting. 05:46:32 Manisha Gundapuneedi: I understand. It's up to you. That's why you need our inputs or we don't you don't need us. We can close it. Okay. Okay. Yeah, Diya Elizabeth: I want to be there. Manisha Gundapuneedi: build that dynamic that you know based on a JSON it is dynamically rendering that you including like CTAs connecting up the CTAs to the next pages things like there is no concept of a next page everything is a single canvas we can approach it that way maybe then based on what they actually say we can red decide. Gopal Gottumukkala: Maybe single canvas is good to start but we have to keep it open. Certain point of time we might become limited because in one Manisha Gundapuneedi: Yeah. Gopal Gottumukkala: single card we might need to produce so much that might not work in certain cases natural drill downs or something and and Manisha Gundapuneedi: Yeah, Gopal Gottumukkala: then I need to tell the user okay you do this but he needs to know why we are Manisha Gundapuneedi: dude. Gopal Gottumukkala: saying that to do a certain thing so things like that but we can start with that but keeping an open And that there could be a possibility of coming back to this single Manisha Gundapuneedi: Sure. Gopal Gottumukkala: canvas and then ability to go down also. Manisha Gundapuneedi: Sure. Okay. Shall we close? Okay. Gopal Prabhakar dropping off Andy. Thanks Andy. Rajashekar G: Sure, Manisha Gundapuneedi: Thank you. Satyasri Prabhakar Mantripragada: Sure. Thank Manisha Gundapuneedi: So Prabhakar next week you please coordinate and uh at least you know couple of people need Gopal Gottumukkala: Yeah, Satyasri Prabhakar Mantripragada: you. Manisha Gundapuneedi: to focus on this simulator. Thank you. Rajashekar G: sir. Gopal Gottumukkala: some Manisha Gundapuneedi: the link Satyasri Prabhakar Mantripragada: It didn't Manisha Gundapuneedi: recording. Satyasri Prabhakar Mantripragada: be. Manisha Gundapuneedi: Thanks. Thank you. Satyasri Prabhakar Mantripragada: Okay. Manisha Gundapuneedi: My sushi order. My You were listening to all of that. Transcription ended after 05:49:49 This editable transcript was computer generated and might contain errors. People can also change the text after it was created.