Feb 23, 2026 Enterprise brain arch - Transcript 00:00:00 Naveen Puttagunta: is specific to that is okay that that Yeshwanth Reddy Yerraguntla: Okay. Naveen Puttagunta: is uh that that is a function of the uh data source. So you can uh that part you can build it into the agent. Okay. then uh so then your knowledge graph is getting updated right so there is somebody I don't know whether that you would put that into the agent or whether that is part of you know some other agent is building your you know updating your knowledge graph based on the latest uh data changes okay then there is that that uh then the intelligence to uh uh detect a problem Yeshwanth Reddy Yerraguntla: Yeah. Gopal sir not huh? Satyasri Prabhakar Mantripragada: are mut. Gopal Gottumukkala: I'm Yeshwanth Reddy Yerraguntla: Yeah. Yeah. So I don't know if there is any disconnect. Gopal Gottumukkala: dead. Yeshwanth Reddy Yerraguntla: I just wanted to again bridge the gap because you joined late. So from the core uh reactive uh agent to make it proactive. What we thought was there are to be two different uh new elements that need to be added. 00:02:55 Yeshwanth Reddy Yerraguntla: One is this concept of cadence or active pull of data. Who is pulling? I think that is what we are discussing right now. But somehow the whole system needs to refresh its own state of outside world and compare it with the outside world uh sorry compare it with the previous uh state to see if something is worthy of escalation or not. pro to say um a particular uh task in Jira it went from uh QA to back to u development. That is sort of a red flag that ideally from QA it should have gone to done or something but now it moved back because of some issue. based on the previous snapshot it knows that yeah something is worth uh escalating to either the team lead or whoever it is when it comes to me emails the there might need not be any snapshot I'll just pull the latest uh mails from past 1 hour I'll read all the mails and see if there is any negative sentiment if there is anything negative uh let's say some project uh external manager pointed something negative. 00:04:17 Yeshwanth Reddy Yerraguntla: So I'll go search for that project related information in other sources and see is there any correlation or any other information I can attach with this negative source of information to the right person. Right? So this this concept of pulling the data and comparing it is what we we should start thinking of on to we want to go in that direction mainly. Second thing on top of that is the personalization aspect itself that there is a trainer persona who will teach what are good patterns and good uh or antiatterns. When I ask for certain question in Salesforce, there might be several ways of creating a Salesforce query. But for our organization, this is the best way to generate the queries, specific tables key to prefer uh view based querying versus something else based. So this kind of a knowledge for the connector has to continuously evolve based on the questions people are asking. I the agent should not create those queries from scratch. If 100 people have asked that query, ideally speaking, at least 98% of the time, there has to be some cash that it should reuse to come up with the answer quickly. 00:05:49 Gopal Gottumukkala: I understood it Yeshwanth Reddy Yerraguntla: So this evolution of agents own knowledge how people are Gopal Gottumukkala: from Yeshwanth Reddy Yerraguntla: talking how they are expecting the answers to be personal preferences organization preferences all these things is the second aspect that enterprise brain has to uh now also provide Naveen Puttagunta: So um uh Gopal good that you joined and um so I like from a architecture perspective right one of the things that I was u I wanted to discuss is how it is you know how the whole work is split and who does what and all of that. One of the things that you know is um uh would be good I think is see the when we say you know let us take uh Jira as Gopal Gottumukkala: Hey. Naveen Puttagunta: our uh uh example so that it is easy to understand for everybody uh the fact that hey Jira is a a project uh management task management system and then you know you know different companies use it for different purposes also uh then um you know there there are uh entities called projects, there are entities called uh stories um uh epics. 00:07:16 Naveen Puttagunta: Now there is a concept called pro um you know uh velocity, there are sprints. So all of this information should be should come with the agent. So when we're developing that agent, these are all the things that we will um that agent will uh will will u will be uh taught or rather it will come with uh its its own knowledge model or whatever. Uh it it has all of that information. It is it is kind of that self-contained right. So um the fact that oh the projects are probably the most important uh aspect in this uh the task status is uh somewhat much lower down in the priority but the frequency with which the task status changes is uh you know it could change multiple times a day whereas new projects get added maybe uh once a week you know that is the frequency at which those entities change. Um then you know given this priority all of that that kind of information everything about the agent and then uh everything about the data source and how to retrieve that information how uh you know if somebody asks for project velocity you know how to compute it uh all of that should come from uh should be published in the agent itself or rather it it comes with the agent and later on also in the trainer mode when we teach it a new skill that Hey uh by the way when somebody asks for um um you know 00:08:50 Naveen Puttagunta: burndown uh the burndown rate means this okay this is how you compute it should then be encompassed in the agent itself rather than somewhere in the main uh you know in the core platform so that that agent is everything about that data source really understanding that is how I see it. Okay. Then I was also suggesting that there is okay and and this could be done in both ways. Uh I was thinking that there is a refresher agent that would uh based on the enterprise uh uh that particular enterprises uh priorities and preferences. uh the it could go after each uh data each agent and say uh give me the latest data right but but perhaps Yashwant is suggesting that actually that um you know that frequency update can be built into each agent that uh it knows uh what is the uh frequency at which things need to be pushed out and so it supplies that and gives it to an agent that kind of rebuilds the knowledge graph or whatever. And perhaps there is another agent that watches for um insights uh into oh this particular task regressed uh rather than progressing it it went back in this thing that is an insight. 00:10:21 Naveen Puttagunta: Now correlated with that is some other piece of information that could come from a different system. So these insights are then correlated and then it could mean an escalation or a notification. So each insight may not necessarily immediately trigger an ins escalation or insul notification but uh some insights correlated together can can uh mean an escalation or notification. So something along those lines, right? So I mean it's not just for notifications or escalations. I'm just mind you I'm just saying overall the system if we look at that system architecture should we uh you know how should we structure it so that these things are cleanly separated Gopal Gottumukkala: Yeah, Naveen Puttagunta: out Gopal Gottumukkala: got it. Now this is almost like the except frequency going inside frequency of update going inside the agent. This is what we discussed earlier also right now. Naveen Puttagunta: Yes. Gopal Gottumukkala: So nothing much changed since then but basically what we are saying is Naveen Puttagunta: Yes. Gopal Gottumukkala: there could be many such agents doing different jobs on different sources. Naveen Puttagunta: Yeah. 00:11:29 Gopal Gottumukkala: So basically at a very high level a source is associated to an agent everything and anything related to that particular source that agent has the knowledge and also the task that it is supposed to do that we are calling it as agent and then there is another level of uh agent or LLM which can understand the language and then figure out which agents to I mean uh what to be asked to what agents so that it can it can answer at a higher level. Naveen Puttagunta: Yes. Uh yeah, Gopal Gottumukkala: Um Naveen Puttagunta: you could you could u so again you know the there is a correlated knowledge graph right. So uh um it is not just a uh request response kind of a thing because that would again that that is the NLSQL right where we don't build anything uh outside the data sources but here we do want to build a correlated I am very specific in saying correlated knowledge graph right I don't know whether you they're embeddings or knowledge graphs but I am saying somehow you are able to correlate different pieces of information and you are able to store it in your knowledge graph or vector know embeddings or whatever. 00:12:58 Naveen Puttagunta: So when in in sometimes if you ask a question perhaps okay um you know that knowledge is already within the system so you can immediately answer or if it needs references it knows where to pull things from Gopal Gottumukkala: Got Naveen Puttagunta: meaning uh yeah I'll just to give a specific example right let us say Gopal Gottumukkala: it. Naveen Puttagunta: uh you know in an email uh there is a reference reference to um you know uh Tata Steel. Now you know the orchestration if I call it an orchestration engine you know does it implicitly know that hey Tatast steel is also a um uh prospect in the sales pipeline because the there is something in the um in in Salesforce about it and by the way it has progressed to such a point that there is already a P project underway because you know you see it in Coutilia and there is a correspond responding project in zera where you have you are kicking off something that where you are actually specifically building it. So it knows that oh all of these places also have information about Tatastel. 00:14:14 Naveen Puttagunta: It doesn't have to go query each one in turn to say do you know about Tatastel? Do you know about Tatastel? Right? It knows that oh Tatastel has all of these references in these because the status of Tatastel in the thing is at a prospect and it is at a discovery level. Um there is coutilia it has progressed to a point where we have actually allocated resources the fact that uh oh there is a specific active project underway to deliver something for and and then the project is like 25% done. So it knows at least this level of information in the knowledge graph. Then you want to very very specifically ask a particular question that needs references from the sources. Then uh it will go ask those relative agents the respective agents those information pulls that together and gives you an answer. Gopal Gottumukkala: This is clear. Nin. Uh say assume while it is inferring say this agent Naveen Puttagunta: Okay. Gopal Gottumukkala: knows when to pull the information right. It pulls information. 00:15:26 Gopal Gottumukkala: It figured out could you go to that circle over circle right? It figured out a negative pattern. Yeshwanth Reddy Yerraguntla: Yeah. Gopal Gottumukkala: I mean it identified that there is something that is negative and then how does that initiate because this agent has only Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: Okay. Gopal Gottumukkala: a knowledge about this but of course it has a correlated knowledge still utilizing that correlated knowledge doesn't make any sense. I mean that ability agent to have it is a little too much of a distributed probably can this inform the parent to initiate certain process or I'm just throwing out why the entire Yeshwanth Reddy Yerraguntla: H. Gopal Gottumukkala: solution capability has to be with every agent is what my question was ideally there is some bigger model has everything it knows uncertain Yeshwanth Reddy Yerraguntla: Yeah. Gopal Gottumukkala: events what to In order to do that or figure out it is going to use all the agents that is that are available for that. Now because the data pool you know study is happening by the agent. It is not because somebody called it. 00:16:38 Gopal Gottumukkala: It is just doing its own inference and it found something then or maybe it would have been already discussed but I I think I missed that aspect. Yeshwanth Reddy Yerraguntla: Le I was also actually talking on mute. Gopal Gottumukkala: Yes. Yeshwanth Reddy Yerraguntla: Can you hear me? Gopal Gottumukkala: Okay. Naveen Puttagunta: Yeah. Yeah. Yeshwanth Reddy Yerraguntla: I yeah I wanted to interject and say this thing but goals are also told similar Naveen Puttagunta: Yeah. Gopal Gottumukkala: Now Yeshwanth Reddy Yerraguntla: something similar like what I was thinking was every agent has the power to escalate it to the uh main agent Chanaka right Chanaka then decides based on the escalation okay I think I'll have to fetch certain information from every other relevant agent and it knows which of the you know three agents or which of the 10 agents to ping and get that information is what I was also thinking when I was creating this diagram because a this a zaka is the source of that correlation is how I thought but think we Gopal Gottumukkala: Yeah, Yeshwanth Reddy Yerraguntla: can continue this yeah I mean 00:17:48 Gopal Gottumukkala: then we are good. Naveen Puttagunta: Yeah. So it Yeah. Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: earlier I was uh imagining that the each individual agent like the Jira agent is just a a data retriever and uh it you know it knows about the entities and the relative priorities and all of that. Um but you know based on what Gopalaru and you Ashwant you are also suggesting it doesn't have to be that dumb so you what we're saying I mean I initially did not imagine Yeshwanth Reddy Yerraguntla: Honest. Naveen Puttagunta: that when it um you know because based on its uh cadence data refresh cadence it is going in and seeing the uh latest tasks and it figured out that 10 tasks uh 10 tasks in a single project uh uh regressed Right. They went from inqaeda back to production back to development which means actually so earlier I was thinking that it will not either uh um uh it will not draw any inference from it but you know I can easily say that it doesn't have to be that dumb it knows that oh there are 10 negative aspects. 00:19:03 Naveen Puttagunta: So um so now again does it escalate or does it just um record the 10 insights that uh oh this particular task regressed this particular task regressed this particular these are notable insights right so I can see that actually that is better coming from the agent because that is again specific to that data source. So what is concerning in a data source and what is the relative priority of that concern is obviously specific to each data source right because that cannot be translated across. So I can see that the data the agent also understands what is important what is not important what is an escal what is an escalation level worthy uh escalation worthy level uh uh insight versus oh just an interesting insight all of that it will I think it can process and then the insights are then posted to a uh a common uh insight uh uh insight place that again and I you know that uh you call it a z or something else then processes these insights and then Yeshwanth Reddy Yerraguntla: Okay. Naveen Puttagunta: understands okay there is something that is really uh worthy of escalating it to a specific person right because again different people are uh have you 00:20:28 Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: know it's not like it's a companywide escalation it is just an escalation for one person uh and that person needs to So uh something like that right because that is where the personalization comes in. Yeshwanth Reddy Yerraguntla: Hello. Naveen Puttagunta: I mean it determines who needs to be as credited but or even if there is no such designated person there now it is a you know there is a P 0 uh event right so there there is so then there is a centralized you know all of those insights are correlated and then you know compound insights are formed and then uh someone else will then start take the responsibility of Yeshwanth Reddy Yerraguntla: No. Naveen Puttagunta: okay distributing notifying um you know and then obviously this is the Yeshwanth Reddy Yerraguntla: No. Naveen Puttagunta: place that when somebody logs in the first thing you pull from is oh these compound insights are what is exposed. Does that kind of make sense? Yeshwanth Reddy Yerraguntla: Okay. Naveen Puttagunta: I mean uh you know uh that kind of thinking of okay the agent knows some level of this but then the um correlation and all of that happens outside and on and it's not in obviously in the agent 00:21:46 Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: itself. Yeshwanth Reddy Yerraguntla: Orchestrator's additional responsibility is to correlate these additional informations and uh decide uh on everybody's dashboard what I need to pump. Yeah, I mean that so yeah I mean Chanaka works both ways now. Not just to decide, hey, user is asking me something, I'll figure out whom to ask, but I've been receiving so many informations and insights from every agent. I will decide which users are the right people to distribute these informations to. Naveen Puttagunta: And first of all as new insights are being added and in fact is the data being refreshed is there something called like oh you I have to now update my knowledge graph with the new data and the new insights. Yeshwanth Reddy Yerraguntla: I haven't written that down but it I think it should do that. Chanaka itself has its own knowledge graph now dedicated uh KB which is different from the knowledge graph of every individual agent. This Naveen Puttagunta: And I'm not even sure that you do you need an individual LLM knowledge graph. 00:23:02 Yeshwanth Reddy Yerraguntla: we are calling it KB but it's just knowledge end of the day these snapshots right for every agent how the things are evolving is what I'm what I was calling as knowledge graph but yeah end of the day yeah it's just memory for the agent to Naveen Puttagunta: I see. I see. Yeshwanth Reddy Yerraguntla: know how things are evolving I am saying there is now an additional new Naveen Puttagunta: H Yeshwanth Reddy Yerraguntla: memory only for Chanaka whose only job is to synthesize information from all the downstream agents and understand at an organization level or at a very high level across domains what is happening. Naveen Puttagunta: Okay. Yes. And and basically see um fundamentally just like how we did it in optin, right? We always have to ensure I mean make sure that the enterprise brain core platform whether you call it the core at the core platform level or at the agentic level um all of them have the ability to learn new skills on the fly in a trainer mode right and then they can add it to their own 00:24:10 Yeshwanth Reddy Yerraguntla: Yeah, Naveen Puttagunta: um you know uh set of skills so that the skills can be updated Yeshwanth Reddy Yerraguntla: got it. Naveen Puttagunta: dynamically. right? We don't have to. Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: And so uh again the big I mean for me the biggest uh advantage if in that kind of a model is when we go to a new data source somebody can sit down with it on trainer mode and just teach it like how we train a a new hire right and it's it it is not Yeshwanth Reddy Yerraguntla: Good. Naveen Puttagunta: coding work but it is more prompting work really understanding and then you know so a subject you could potentially put a subject matter expert in front of it and say okay now develop that agent. So our raw agent basically is nothing but a connector to that system. It just understands that system which understands the u language and then after that it it can be trained uh um either through just uploading a document or just Yeshwanth Reddy Yerraguntla: correct. Naveen Puttagunta: prompting. Yeshwanth Reddy Yerraguntla: Yeah. 00:25:16 Yeshwanth Reddy Yerraguntla: Yeah. So to teach every agent its own set of skills uh is definitely a common Naveen Puttagunta: Okay. Yeshwanth Reddy Yerraguntla: pattern. Yeah. Naveen Puttagunta: H. Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: Okay. Yeshwanth Reddy Yerraguntla: Um I mean I didn't we didn't come to that part but yeah like agents memory did contain what are those trained patterns and what is the organization knowledge it's working with. Uh so some trainer persona will come into picture. Um and that that trainer teaching the agent at a chatbot at sorry at a chat interface level and that percolating from Chanaka to the agent to the agent skill is something we have to uh uh take into account. Naveen Puttagunta: Yeah. So, Gopalo any um any comments on this uh the refresh uh then you know rebuilding knowledge graph insights things like that. Gopal Gottumukkala: It was in bad. I was in a bad patch. Naveen Puttagunta: Okay. Okay. Okay. Okay. No problem. So then I'm saying and did you catch some of this and that we're talking about this and agent it's clear that we all agree that the agent will have its own uh uh skill sets. 00:27:15 Naveen Puttagunta: It'll all its own uh idea of all the what entities are available, what are the relative priorities, what is the cadence at which you have to uh you know push that data out based on the priority uh you know what does something mean etc. And it also has enough knowledge and skill to understand okay if this happens that means it's uh positive or negative or how critical something is uh relative to that s data source then basically it is able to do that and then uh push that out to a a central repository where those insights are then stored and also it it can add data to the central knowledge graph also so that the central knowledge graph can the orchestrator can then correlate uh the different items and then decide oh this is what needs to be done. Does that make sense? Gopal Gottumukkala: Yeah, that is clear. Naveen Puttagunta: So what in in you know what do you think that this is for both Yashwan and Gopal. Um see when you know suppose some task in Jira has changed Gopal Gottumukkala: I 00:28:37 Naveen Puttagunta: status. Okay. Do you think that this task changed status is something that Gopal Gottumukkala: think Naveen Puttagunta: you would um you would uh you know does the agent update its own knowledge graph or the uh the central knowledge graph is also updated with that is that something like that at that importance level do we uh bring that into knowledge graph or not uh not required. Gopal Gottumukkala: we should be the right person to answer nin but in general if we have a representation we should but what actually we are having in the knowledge is important if every task is there then it is a kind of too much of information I mean task status but I'm not sure Yeshwanth Reddy Yerraguntla: So the way I want to put it is uh uh yeah I mean I resonate with what Gopasar is saying that we should not really uh try to if if we try to do these things essentially we are trying to duplicate the whole Jira which we should try to avoid we should try to strike that balance that said if a particular user is a good case where Navana whatever he's suggesting actually comes into picture is when a particular user is interested in those things. 00:30:06 Yeshwanth Reddy Yerraguntla: So as a team lead I am interested in uh uh planning for my next sprint. I asked my team to come up with uh the task for the sprint. They have come up with them. One of them one of the field in those task is acceptance criteria. it now if the agent knows that I as a team lead am interested in I'm about to do sprint management and I sort of logged on the sprint for Monday but Friday some description changed that actually makes sense to escalate that actually makes sense to put it put as part of the central knowledge graph for the team it for that few days only like for a couple of days it is important to show that information that uh you despite the freeze something changed you should probably take a look. Does that make sense? Naveen Puttagunta: So you're saying you know uh if we specifically take Jira as an example the fact that uh a a task changed its status uh on the surface it looks like too trivial to add to the central knowledge graph because that is just duplicating Jira. 00:31:26 Naveen Puttagunta: But for example um you know when we are when the project is Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: uh in UAT I mean like um yeah as you said after code freeze you know 10 tasks change their status from um uh innuat back to uh regressions right so that is a significant enough uh so we are saying that only significant insights are then forward forwarded to the central knowledge graph. Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: And what is significant is something that you teach the either you teach the agent or that that that information the agent has to understand what is significant in this particular data source. Yeshwanth Reddy Yerraguntla: Correct. Yeah. And it it is time sensitive also. Naveen Puttagunta: So Yeshwanth Reddy Yerraguntla: That's what I'm trying to say. Naveen Puttagunta: yeah, so all of that is in the agent. So right we are saying that uh uh and I can easily see that uh um um Yeshuant see in a different context let us say somebody says no no I want a real Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: realtime monitoring for my Jira okay uh uh because you know working with Jira is so hard I want a real time monitoring dashboard for Jira And that is all I want enterprise brain for. 00:32:58 Yeshwanth Reddy Yerraguntla: Mhm. Naveen Puttagunta: So in that context even the fact that one task changed its status even in a happy path might you might want to add it to the knowledge graph. Yeshwanth Reddy Yerraguntla: Correct. For that Naveen Puttagunta: So uh um not just for the I mean yeah see again I come back to Yeshwanth Reddy Yerraguntla: user Naveen Puttagunta: knowledge graph saying knowledge graph is interesting that you say you know for that user shouldn't it be the source it's not like what we pull from the source uh does it have to be specific to the individual users? Yeshwanth Reddy Yerraguntla: what I'm pulling what I'm pulling from the source it has to be user dependent is what I feel Anna because what praakers is interested in uh out of Jira and what I am interested out of Jira are different Now we can't that is where user preferences comes into picture. That is where Chanaka's understanding of who it is dealing with comes into picture. But at a data source level itself we should not micromanage it for every user because then we are again doing a multiplicative uh complex task. 00:34:17 Yeshwanth Reddy Yerraguntla: Now for if there are n connectors and m users now we need to deal with m into n combinations. Naveen Puttagunta: Yeah. Yeshwanth Reddy Yerraguntla: Whereas if Chanaka is aware of user's interests, Naveen Puttagunta: Um Yeshwanth Reddy Yerraguntla: Chanaka can dispatch those necessary instructions to the data. So that just for that user, this data agent is now acting in a slightly different way. Naveen Puttagunta: okay so it's almost like we are saying normally oh every task and every status is not important but uh you're saying that hey for prabhakar um you know because he is he has to do hourly tracking of one particular project because it's so critical there he does want that ability so only for that project you do you do that uh data Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: push but it is not I mean once it is once you decide Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: to push it and add to the knowledge graph it is the universal knowledge graph I mean you know um if someone else is also interested in the same project, you're not going to pull that data twice, 00:35:29 Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: right? Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: Because it's the same data. Yeshwanth Reddy Yerraguntla: That will not happen. Naveen Puttagunta: So knowledge gra it'll just get added to the knowledge Yeshwanth Reddy Yerraguntla: Yeah. Like I'm saying this on the fly but now I'm thinking will it be difficult? Naveen Puttagunta: graph. Yeshwanth Reddy Yerraguntla: Will it not be easy or what are the implications of this kind of a architecture? like from a customization personalization standpoint, this definitely is a much better way to handle the LLM itself because we are taking advantage of it. In that Naveen Puttagunta: Yeah. Yeshwanth Reddy Yerraguntla: case, Gopal Gottumukkala: Sorry to sorry to cut cut your thoughts Yeshwanth Reddy Yerraguntla: I'm also in the dark. Gopal Gottumukkala: but basically we we might be trying to solve problem that agent should solve and what knowledge moral should actually acquire over a period of time. Maybe in my mind we might be combining analytics into but what my Yeshwanth Reddy Yerraguntla: Mhm. Gopal Gottumukkala: understanding is knowledge has nothing to do with the facts. It has to just absorb the facts, create the knowledge and it told me has to retain the knowledge not the 00:36:40 Yeshwanth Reddy Yerraguntla: Okay. Gopal Gottumukkala: facts because if we are see enterprise brand has to produce both. It has to have a knowledge and go and get the analytics or get the analytics and then produce I mean do the uh extrapolations and then give an answer but an individual agent or individual model but should not even try to remember anything the system should remember in the sense if I have a data source that has all the analytics built in assume that I have a data loop which Yeshwanth Reddy Yerraguntla: H. Gopal Gottumukkala: will give me if I you fire a right SQL it will give it will give you the facts knowledge I mean models have to go get different kinds of on their own get try to get different kinds of patterns and remember that this is how this system works so if I consider Yeshwanth Reddy Yerraguntla: Correct. Gopal Gottumukkala: Jira is also source I don't want to certainly remember what is that particular task status that is a runtime question and ability to pull that answer what it should remember is Gopal is Yeshwanth Reddy Yerraguntla: No, but here. 00:37:56 Gopal Gottumukkala: historically lazy and he don't update the tasks or his tasks are long Yeshwanth Reddy Yerraguntla: Okay. Gopal Gottumukkala: pending that kind of knowledge is what model has to remember if he can bifurcate that you know data is available whenever I want I am only acquiring the knowledge then particular task status changed does it need to go to knowledge graph or the knowledge has to go to The knowledge knowledge graph is something yesterday also I got confused here only I'm still debating myself that why do I need to remember anything that is a fact I just need a knowledge out of it something happened I need to bump up my knowledge it is something like you know India's prime minister changed right Yeshwanth Reddy Yerraguntla: H. Gopal Gottumukkala: and yeah obviously there are certain things that it has to remember is entity is so so far I don't have any anybody called prime minister I need to remember that you know country will have prime minister but I certainly don't want to remember at this point who is the prime minister that I will go and get it in an enterprise brain Yes. 00:39:14 Yeshwanth Reddy Yerraguntla: uh yeah I think we should discuss a few more minutes you said to pull the data uh is something which necessarily means it is reactive, isn't it? Gopal Gottumukkala: Huh? Yeshwanth Reddy Yerraguntla: Like Gopal Gottumukkala: See pull the data is just an ask question but I could decide when to pull the Yeshwanth Reddy Yerraguntla: you Gopal Gottumukkala: data. So I am continuously trying to learn I know that that system is changing continuously. Yeshwanth Reddy Yerraguntla: got Gopal Gottumukkala: So J is a little tough for me to digest but if I go and look into a Yeshwanth Reddy Yerraguntla: it. Gopal Gottumukkala: database database is constantly changing or it is changing at a certain frequency. Yeshwanth Reddy Yerraguntla: Okay. Gopal Gottumukkala: So worst case scenario is in a daily basis, Yeshwanth Reddy Yerraguntla: Correct. Gopal Gottumukkala: right? I need to go back and then infer whatever I was looking for Yeshwanth Reddy Yerraguntla: Yeah. Gopal Gottumukkala: in that particular database. I was looking certain n for things, right? I'll go and see what happened to those n aspects. I do it on my own. 00:40:11 Yeshwanth Reddy Yerraguntla: Right. Gopal Gottumukkala: But when a question comes based on that knowledge, I know what to do. And if the question itself is specific, then I have to go back to the source and get the data and present them. But something like uh you know how how I'm doing doing Naveen Puttagunta: self. Gopal Gottumukkala: at sales right for the first level answer my Yeshwanth Reddy Yerraguntla: H. Yeah. Gopal Gottumukkala: knowledge tells you're doing good for the past four years your sales is increasing in past five months also your sales is increasing but if I have to produce the numbers I have to go back and get the numbers I know how to get but I don't have the numbers with me this is what I was thinking I could not translate exactly the same thing onto Jira. But if it is structured data, that is how I would like to remember. My question is suppose Yeshwanth Reddy Yerraguntla: I'm sorry. It's eviding my understanding. Gopal Gottumukkala: uh I don't want to pull you to your understanding but I I would like to put my thoughts so just keep it floating what I'm saying is say take case right how many opportunities are in open state is the 00:41:36 Naveen Puttagunta: I'm Gopal Gottumukkala: first level answer that comes from LLM without even touching the sources Naveen Puttagunta: sorry. Gopal Gottumukkala: typically you know uh you have around 30 opportunities open on an average for the last four months. Yeshwanth Reddy Yerraguntla: Right. Got it. Gopal Gottumukkala: But as of today, Yeshwanth Reddy Yerraguntla: Got it. Gopal Gottumukkala: I want a precise answer, then I'll go back and get it. I cannot remember that because that is so much of remembrance I have to Yeshwanth Reddy Yerraguntla: Okay. Gopal Gottumukkala: do. Yeshwanth Reddy Yerraguntla: too many things to remember. If you start remembering every little thing, Gopal Gottumukkala: Yeah. Yeshwanth Reddy Yerraguntla: uh you're wasting memory, right? You're not really doing any Gopal Gottumukkala: Yeah. Naveen Puttagunta: This Yeshwanth Reddy Yerraguntla: indexing. Naveen Puttagunta: is Gopal Gottumukkala: that that's what my thought process was and and most of the Yeshwanth Reddy Yerraguntla: I see that I think we are aligned at I think we are aligned on that Gopal Gottumukkala: time and now when Yeshwanth Reddy Yerraguntla: part that we should not remember every Yeah. Gopal Gottumukkala: we come to yeah if I come back to zer right the particular task status changed something that I need to infer but do I need to remember the task 1 2 3 4's current status is in progress 00:42:46 Yeshwanth Reddy Yerraguntla: It is worth remembering for that one day is what I think Navin Nana and I are on the same on that page Gopal Gottumukkala: Huh? Yeshwanth Reddy Yerraguntla: because Gopal Gottumukkala: I I I understood I heard that point that much. I agree. There is certain I mean current happenings you better remember but that should not be pushed back into knowledge graph that it will be permanent. That was only the knowledge has to go. Yeshwanth Reddy Yerraguntla: Okay. Gopal Gottumukkala: I I don't know how to put it exactly in the a terminology but that's what my thinking process was and it be total Yeshwanth Reddy Yerraguntla: Definitely point that knowledge graph knowledge graph should not be Naveen Puttagunta: I will Gopal Gottumukkala: wrong. Yeshwanth Reddy Yerraguntla: permanent. Naveen Puttagunta: give. Gopal Gottumukkala: Okay then then I Naveen Puttagunta: Let me uh let me let Can you guys Yeshwanth Reddy Yerraguntla: Yeah, but it's Naveen Puttagunta: hear? Yeshwanth Reddy Yerraguntla: breaking. Naveen Puttagunta: Oh, okay. How about now? Yeshwanth Reddy Yerraguntla: Yeah, Naveen Puttagunta: Still waiting. Yeshwanth Reddy Yerraguntla: better. Naveen Puttagunta: Okay. So, what I you know I'll give different examples, right? 00:43:57 Naveen Puttagunta: So you are watching Salesforce uh to watch the sales pipelines. Let's say we got a uh an interest uh request from uh uh Tatastel. Okay. So Tatastel was lo is logged as a um uh as a new lead. Now uh in Salesforce the agent knows that uh hey the leads are leads the company names of the leads are the one of the most important high priority items. So it immediately pushes that hey we have a new lead uh called Tata steel. So because it is such a high priority uh element I do believe that the knowledge graph has to be updated with hey uh Tatast steel is now a new lead in the pipeline and um you know um so who in Tatast or even like let's say people are also their people are also extremely important right from a business standpoint so that is extremely critical. Let us say the lead value is uh you know $300,000 because the lead value is 300k and for us it's a it's a substantial u lead you know amount the fact that hey actually the lead also came with a high value project is the knowledge I would like to retain but the the second aspect I mean the like 300k can change to 100k 100k can change to 250k. 00:45:45 Naveen Puttagunta: So every time it changes uh it is important but uh less important than the uh identity of Tata steel itself. So let's say uh the uh I'm I'm trying to think of use cases as I go along. So bear with me. So let's say the uh project value was updated from 300K to let's say 100K. Uh so the fact that hey the the value dropped is uh it's not an escalation level but it's an insight that hey uh you know uh the value the project value has been substantially reduced. Now uh you know what is the exact value is the fact that I don't mind going and querying in in dynamic dynamically but the fact that oh okay the project value reduced is a is a knowledge item that uh depending on specific you know uh users preferences and all of that it may be a notification for them. Hey, the Tata sales project value reduced. How long do you remember that? Uh that uh we can debate. Uh but the fact that Tata Steel has been a lead is actually something that I would like to remember forever because you know 3 months later we have another email team that needs to be correlated. 00:47:19 Naveen Puttagunta: Right? So that kind of stuff is what I'm talking about. the the fact that Tatast has been a p, you know, something in the pipeline, I want us to uh um I I want us to immediately uh focus on Yeshwanth Reddy Yerraguntla: One one additional perspective I can add here is Tata steel by itself is not something that entire organization needs to know but people for people like Navin and Nadiva um sorry Pratima Naveen Puttagunta: Sales Yeshwanth Reddy Yerraguntla: the respective salesperson not everyone in the sales team also probably and the team leads assigned team lead and assigned squad. Naveen Puttagunta: I Yeshwanth Reddy Yerraguntla: Now it is important for every one of these persons that in the in their knowledge Naveen Puttagunta: believe Yeshwanth Reddy Yerraguntla: base this is going to be one of the high priority item. So it somehow shaka knows this is a 3 month project. So maybe the item list will go away after 3 months but tentatively pet that for these 10 people in Dwami I have to be on top of what is happening with Tata Steel and immediately put every information 00:48:31 Naveen Puttagunta: That's Yeshwanth Reddy Yerraguntla: necessary in the dashboard in the right fashion for the right person say the Pratima madam need not worry about what's happening in Jira or maybe she is worried in the sense that or the or the Apex it's going properly or not for a Naveen Puttagunta: Correct. Yeshwanth Reddy Yerraguntla: lead. Naveen Puttagunta: for for Tatil especially. Yeshwanth Reddy Yerraguntla: Correct. For a lead, it's slightly more higher resolution. Nonetheless, Chanakya has to remember as much as possible about artist for these 10 people Naveen Puttagunta: Okay. Yeshwanth Reddy Yerraguntla: is something it has to do and that is something that has to be across all the database sources which means we should not put the burden on Jira data source or JA Jira agent to remember anything about artistry. This is how I would look at why or how knowledge base should work for Chanaka. Naveen Puttagunta: Yeah. And by the way, why do you say uh you know or Tata Steel is in the pipeline is important uh for these 10 people's knowledge base. The knowledge base is central, right? 00:49:57 Naveen Puttagunta: It's not uh divide sliced up by people. It is common for Yeshwanth Reddy Yerraguntla: Knowledge base is central but knowledge base also contains for every person what are they working Naveen Puttagunta: everybody. Yeshwanth Reddy Yerraguntla: on that is how you are able to slice the knowledge base and sort of do a dynamic filtering on the fly I'm not saying for pratimagaru these 10 items I'll remember I will say for Pratimagaru Tatastel is important. Here are the stuff for relevant to Tatastel in the same knowledge base in a different format. Naveen Puttagunta: I I Okay, let me try to put it slightly differently. You you are I hope you are saying that you know that for Pratima um Anything to do with the sales pipeline is important. Uh something you know anything to do with projects uh getting going off track is important because it it will impact customer relationships. Anything impacting customer relationships is important for Pratima. Yeshwanth Reddy Yerraguntla: H. Naveen Puttagunta: I hope that is how we remember right not specific entities because we you don't know that that there is even a new entity that will come but the fact that a 00:51:13 Yeshwanth Reddy Yerraguntla: Got it. Naveen Puttagunta: new entity will come is also important if it falls under those criterias Yeshwanth Reddy Yerraguntla: Right. Yeah. I was focusing to um like tunnel vision or you Naveen Puttagunta: right Yeshwanth Reddy Yerraguntla: you properly put it. That's how it should Naveen Puttagunta: Okay. Yeshwanth Reddy Yerraguntla: work. Naveen Puttagunta: Okay. So, so the so the the thing is if I got a uh let us say I mean so I am trying to you know talk about an edge case right um so let us say Tatast steel didn't even get added into Salesforce. Yeshwanth Reddy Yerraguntla: Okay. Naveen Puttagunta: Okay. Yeshwanth Reddy Yerraguntla: Can you repeat that? Naveen Puttagunta: Um uh um the Tata Steel didn't even get added into Yeshwanth Reddy Yerraguntla: Sorry. Naveen Puttagunta: Salesforce as a lead. Somebody who supposed to enter it as a data entry thing, they didn't do it. But you know, emails are starting to fly and uh you know there's a chat uh specifically created for Tata Steel and that is happening. So it is clear to so in the end I want it to be clear to the orchestrator that hey Tata still actually became a you know we are actually working on it. 00:52:31 Naveen Puttagunta: So either it is a customer project or it is a P. However, something didn't even get into uh Salesforce right that part of oh something didn't get into Salesforce is a action item for the sales ops person. But the fact that we are now recognizing that Tatast steel is also a uh you know is a lead or a prospect or a customer because there are projects about it and there is a talk about it that still needs to get into Pratima's uh u immediate uh uh you know list like interest list that hey there is a new uh project that people have started working on. Okay. Yeshwanth Reddy Yerraguntla: H. Naveen Puttagunta: Now the then then why is it I mean uh the fact that it was not entered and all of that that probably will you know will come as some escalation for somebody Yeshwanth Reddy Yerraguntla: All right. Naveen Puttagunta: else. So that that is also what we're looking at. I mean that's the kind of thing that we want to enable. Yeshwanth Reddy Yerraguntla: I'm good. Naveen Puttagunta: Uh okay. 00:53:50 Naveen Puttagunta: What's uh Yeshwanth Reddy Yerraguntla: I mean next steps wise uh I can sit with Raja and Manisha and clean this up. Uh Gopal sir um do you have any suggestions or direction you want Naveen Puttagunta: Okay. Yeshwanth Reddy Yerraguntla: me to take? Like I am just thinking we'll have a another round of discussion with another diagram but that probably might be a bad idea. Naveen Puttagunta: So he's talking I think he's he's just summar. Yeshwanth Reddy Yerraguntla: Okay. Naveen Puttagunta: So hang on. Yeshwanth Reddy Yerraguntla: I think to have the ability that this now taste is in the pipeline Naveen Puttagunta: Let Yeshwanth Reddy Yerraguntla: but someone didn't do the right necessary information like project Naveen Puttagunta: me Yeshwanth Reddy Yerraguntla: owner that should be escalated to you or um to Salesforce something was not updated that should get escalated to Pratma at the same time madam can decide uh Naveen Puttagunta: find nothing. Yeshwanth Reddy Yerraguntla: this is the job of can I'm just saying can this is the job of kamill let him take care of it at which point enterprise brace uh enterpine brain enterprise brain also knows to send a notification to camel because end of the day this is this this engine knows how to send notifications should also be worth considering is what I felt 00:55:33 Naveen Puttagunta: Right. Yeah. Yeah. See the the fact that um you know Tata Steel didn't get entered to Salesforce would never be detected by the Salesforce agent right it's Yeshwanth Reddy Yerraguntla: Got it. Naveen Puttagunta: actually the orchestrator which will uh which I don't know how you will detect it but um hey uh you know we're starting to see meetings and uh email chatter and spaces chatter about Tatast But uh you know and the summarization of all of that is that looks like it's actually we've been talking to them about a P but it's not in the salesforce right because there is no correlation with Salesforce that uh the sales it's actually a valid lead so okay some that I I wonder how that orchestration will will work but that is what we're trying to enable also but that's okay I am Okay, Yeshwanth Reddy Yerraguntla: I can suggest we already know how to do it. This is a skill now. This is a new project onboarding skill for Chanaka. Naveen Puttagunta: good. Yeshwanth Reddy Yerraguntla: So it should kick off that skill where the information is present project manager project owner things so and so things need to be present in Salesforce somehow stuff need to show up on Jira like these are all the checklist items that are part of that skill now and it will continue to walk up with everyone until all those items are ticked in it in its own knowledge base attached to Tatast 00:57:11 Naveen Puttagunta: So as as next steps, right? Yeshwanth Reddy Yerraguntla: Interesting. Naveen Puttagunta: So what I think you need to do Yeshuant is um Prabhakar can you just watch out for V and whenever he's back I think he's off uh he's been on leave um can you track whenever he's back we'll also get V to work with the the two teams right uh so and then bring him into this loop one of the things is first drop this architecture or whatever so that it is documented and I then need Yeshwanth Reddy Yerraguntla: Okay. Naveen Puttagunta: it to be uh implementable steps for either Manisha's team or Raja's team right like so what does this mean so even if the full architecture is not Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: clear yet if some part of the let's say that agent framework if we say hey let's take a you know basic agent uh uh structure and these are the things that each agent will need to publish. So let's say entities uh prioritization uh period you know refresh rate you know these are the kinds of uh things that the entities will need to um uh you know the the agents will need to publish that needs to be plugged in you know then each agent will have its own skills each agent will have its own knowledge base. 00:58:36 Naveen Puttagunta: uh so you know if that structure is clear and if that architecture is clear then um let one team take it and and run with it and start building that agent because you don't want to uh just to do architectural discussion for two weeks and then we have not built anything right so even if that agent architecture evolves we can certainly make that change uh after that Yeshwanth Reddy Yerraguntla: H. Naveen Puttagunta: right let somebody write the technical spec for it and then go build that agent Yeshwanth Reddy Yerraguntla: Okay. Naveen Puttagunta: Yeah. Uh that that and then have that uh clarity of what that uh Yeshwanth Reddy Yerraguntla: Sure. Naveen Puttagunta: architecture I mean like that design look component design looks like right where is the uh how does it publish the metadata how does it uh you know take uh how does it consume preferences um Rajashekar G: Uh sir I have one doue here like uh on Saturday we have discussed about that arch type Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: simulation those action action items right so is it the same thing we are going to pick it up both are same or we are dealing that apart and this 00:59:42 Naveen Puttagunta: H Rajashekar G: apart Naveen Puttagunta: the the completely different Rajik that simulation is to kind of simulate how a UI orchestrator AI first UI orchestrator can uh you know dynamically change the UIs. Rajashekar G: Okay. Okay. Got it. So this is completely another next Naveen Puttagunta: Okay. Uh this is uh enterprise brain. Rajashekar G: features. Naveen Puttagunta: How does that agent how does this agent architecture work where we are able to do different Rajashekar G: Yeah. Naveen Puttagunta: things with uh different systems. Rajashekar G: Okay. So got it. Naveen Puttagunta: Okay. Um, Rajashekar G: Yeah. Yeshwanth Reddy Yerraguntla: Rajik sorry just to tell that like Naveen Puttagunta: so Yeshwanth Reddy Yerraguntla: that UI how it the whole thing works I just uh highlighted in the yellow rectangle right that this much is the orchestration if I'm not Oh, I Rajashekar G: Not user will be there. Naveen Puttagunta: Yeah. Are Yeshwanth Reddy Yerraguntla: know. Rajashekar G: It directly need to on while user landing itself it should come up with uh some kind of Naveen Puttagunta: there Rajashekar G: a dashboard. 01:00:51 Yeshwanth Reddy Yerraguntla: Correct. Naveen Puttagunta: the the summarization of that day discussion Yashwant is that there is a framework behavioral pattern framework that the uh the design team has is going to finalize it's not there I mean it's not done yet but it is they're going to come up with it that frame framework basically describes you know what type of people are there and based on their behavior on how they behave in the system you better uh you know you can categorize them into hey there are this type there are this type etc so somebody is watching the user behavior and categorizing updating their categorization in real time so that is let's say one task aspect of it then given that okay there is a you know based on the user behavior and then there are behavioral pattern frameworks that will say okay if this is this kind of a user they want information presented to them in this fashion first they want to be told the summary and then analytics or analytics versus summary etc. So there is a uh UI orchestrator that understands all of Rajashekar G: Okay. 01:02:00 Naveen Puttagunta: this. All right. And and this framework is can be fed in just like with a series of frags or in a document it can be fed in. So it can always be changed dynamically. Okay. Yeshwanth Reddy Yerraguntla: got. Naveen Puttagunta: So nothing is hardwired in right this is learned behavior and it can be unlearned and relearned Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: right. So then the UI orchestrator will consume this and based on the user who has come in and what is the context current context what is the what what does it need to highlight based on its latest knowledge graph all of that it makes a decision as to how to uh you know what data to present and then it also has a you know uh visualization framework for okay for this kind of data for this kind of user the best visual ization pattern is a bar chart for this kind of a data for this particular kind of a user. Uh you know you need to actually show a network graph. Uh so that kind of thing. 01:02:58 Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: So it will then okay what is the next step? What is the drill down? What is this thing? All of that is or you know can be formulinated by the orchestrator UI orchestrator and then Yeshwanth Reddy Yerraguntla: Mhm. Naveen Puttagunta: UI orchestrator simply sends those set of instructions and data source connections to the front end and the front end then uh uh dynamically can you know can render those widgets and render those uh elements. Yeshwanth Reddy Yerraguntla: Got Naveen Puttagunta: So UI has so so what I asked uh uh you know the team to do Yeshwanth Reddy Yerraguntla: it. Naveen Puttagunta: is hey to prove that point I just want a simulator all the simulator is doing is you know you ingest that framework okay and you simulate a series of scenarios uh simulate means you give it a series of input scenarios right okay currently uh this kind of a user CXO user is joining and he's typically a rapid decision maker. Um and currently there there is this thing this is happening in the system. Then the simulator should run through its framework and then say oh here is how the UI should look and it generates the UI description and this widget this widget this widget. 01:04:16 Naveen Puttagunta: So for me what uh the simulator is basically almost like a chat GTP wrapper right like you you use the basic LLM you give it the series of instructions and say based on that oh if this is the user what would we suggest the UI to be and it will suggest the UI right so I so I wanted that simulator to prove that we can actually do the dynamic UI and it will make sense Yeshwanth Reddy Yerraguntla: got Rajar in that case Rajashekar G: Yeah. Yeshwanth Reddy Yerraguntla: anyway it's a I mean yeah it is a different activity But Rajashekar G: Yes. Yes. Yeshwanth Reddy Yerraguntla: the question is should you prioritize that or this? Naveen Puttagunta: Anyway, Yeshwanth Reddy Yerraguntla: Uh I don't know from what I sense it should not take too much time around 4 hours of chatting with chip it Naveen Puttagunta: resolution. Yeshwanth Reddy Yerraguntla: is being expected Rajashekar G: Yes or no? Yeshwanth Reddy Yerraguntla: right. Naveen Puttagunta: She Rajashekar G: I think uh first of all we need to fix a plan like the uh task breakdown I'm still doing and Naveen Puttagunta: opens 01:05:24 Rajashekar G: uh how we need to present like just chart gpt we are going to simulate the UI or taking or Naveen Puttagunta: up. Rajashekar G: something interactive model we will be giving some inputs and generate something So clarity we Yeshwanth Reddy Yerraguntla: That might be we'll have that Rajashekar G: can start. Yeshwanth Reddy Yerraguntla: discussion but yeah I we interrupted you a between the questions. So continue. What were you asking? Rajashekar G: Uh that was the same thing like if uh that is another thing we are dealing and this is another thing we are dealing Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: with. Yeshwanth Reddy Yerraguntla: I mean render key just that we have to deliver a Rajashekar G: Um Yeshwanth Reddy Yerraguntla: timeline. Rajashekar G: yes. Yeshwanth Reddy Yerraguntla: Yeah. Now from what you're saying is while the architecture still takes gets crystallized what things that are anyway necessary to be built let's start building them Naveen Puttagunta: Yes. Yes. because yeah you can't wait for I mean because see first of all that resource Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: allocation and all of that I you know um Tavakar can work with both Manisha and Raja and but they need you know what are the things 01:06:36 Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: to be done in a list right so they can prioritize and build so for me Yeshwanth Reddy Yerraguntla: Agree. Yeah. Naveen Puttagunta: okay I definitely want to make progress on Oracle unifier because that is a critical p item for us to win a project. Okay. At the same time, you know, Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: I don't want to do it in a throwaway fashion where we say, "Oh, this nothing is useful." So, might as well build it in the agent framework. Okay? Uh and you know, whatever we're proposing. So, give them a basic agent framework and make sure that uh then the platform core is able to absorb that basic agent framework. Right? That should be the first order of priority. Second order priority is okay in that agent you know what what is clear to us is okay agent will Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: encompass its own uh data source you know the the entity knowledge you know the skills that it needs to work with just that purely data source no correlation but purely skills that it it understands all of that build it into that agent and so let's develop that Oracle unifier agent with that framework okay um then it will that 01:07:46 Yeshwanth Reddy Yerraguntla: You got it. Naveen Puttagunta: will progress then all the skill learning you know can we encrypt that that anyway we discussed okay that that could also be one task so in my mind you know if if one team is progressing on that at the end of that exercise we will have an Oracle unifier agent in the structure that we want in the agentic system that we want now the core uh can u because they have taken Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: care of changing the core to absorb the new agentic model that'll also uh you know give it to the same team Yeshwanth Reddy Yerraguntla: Mhm. Naveen Puttagunta: don't give it to the other team right then the other team can work on let's say the the U you know so right now the one of the critical items is hey can we do this dynamic UI right so they can start looking at that dynamic UI uh maybe the other things to look at is Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: okay then work on that orchestrator work on that mechanism where okay we're refreshing data uh you know with this uh frequency and based on 01:08:51 Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: that how you know how do we uh absorb those insights correlate insights and then uh raise higher level insights for example I'm calling Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: everything insights right and some insights can turn out to be notifications some insights can turn out to Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: be uh escalations etc but everything is an insight so you process the multiple insights from multiple systems and then correlate and then build compound insights and in in the process of building a compound insight. If you have to go uh uh fetch new facts from the system to add to your insight even it's fine because you know agents support that architecture right so I so that is what I would focus the the one of the teams on is to basically enhance that uh core orchestrator and how do you do Yeshwanth Reddy Yerraguntla: got it. Yeah, idea. Cool. Naveen Puttagunta: And this agentic uh uh framework that we're building this uh connects between different agents, right? You know, we're talking about the core pulling data from this agent and all of that. 01:10:08 Yeshwanth Reddy Yerraguntla: Yeah. Naveen Puttagunta: Perhaps we can do that in a I don't know I mean I don't know if uh you would uh put it in on an MCP framework or a different this but if we make it standard then we can really claim that you know hey we have a Yeshwanth Reddy Yerraguntla: Okay. Naveen Puttagunta: full u you know full-fledged framework that can interoperate very easily it can interoperate with other systems and vectors Yeshwanth Reddy Yerraguntla: Okay. to compatibility with the chanaka and outside Naveen Puttagunta: but I yeah that is something that you Yeshwanth Reddy Yerraguntla: world Naveen Puttagunta: can uh look Yeshwanth Reddy Yerraguntla: makes sense. Naveen Puttagunta: into. Yeshwanth Reddy Yerraguntla: Yeah, I agree. To some extent we are already able to do this. Every connector we write it has its own standalone UI. We can just attach an MCP server to it. Those are already like relatively lightweight tasks. Naveen Puttagunta: Okay, Yeshwanth Reddy Yerraguntla: Yeah, Naveen Puttagunta: understood. Yeshwanth Reddy Yerraguntla: sure enough. We'll take it from Naveen Puttagunta: All right. 01:11:17 Naveen Puttagunta: Okay, sounds good. Yeshwanth Reddy Yerraguntla: Okay. Uh yeah, goals are probably Yeah. Naveen Puttagunta: I think I don't think is Yeshwanth Reddy Yerraguntla: Yeah. I mean whenever Yeah. Naveen Puttagunta: uh Yeshwanth Reddy Yerraguntla: like I'll continue discussion with Manisha and Rasher here itself yeah that's what I was Naveen Puttagunta: okay so should I just drop off just but end of the day have a very Yeshwanth Reddy Yerraguntla: saying Naveen Puttagunta: specific execution plan Manisha Rajar for the different aspects have an execution Rajashekar G: Show us it. Yeshwanth Reddy Yerraguntla: okay Gopal Gottumukkala: I am also dropping the snowing. Yeshwanth Reddy Yerraguntla: thanks Rajashekar G: Yeah. Yeshwanth Reddy Yerraguntla: Right. Um yeah, Manisha maina first thing uh you are aware of this thing right? Oracle Unifier we have to build in the next uh I don't know the deadline but you have to start building Manisha Gundapuneedi: Okay. Yeshwanth Reddy Yerraguntla: it. Manisha Gundapuneedi: Now Rajashekar G: uh the KT is like only enterprise brain ent how it works what is the code Yeshwanth Reddy Yerraguntla: Okay, Manisha Gundapuneedi: Oracle. Yeshwanth Reddy Yerraguntla: little bit turn. 01:13:12 Manisha Gundapuneedi: Wave Yeshwanth Reddy Yerraguntla: So first thing I will suggest is Manisha enterprise Manisha Gundapuneedi: second. Yeshwanth Reddy Yerraguntla: brain Rajashekar G: Serious. Yeshwanth Reddy Yerraguntla: most of our framework All you have to do is pull the correct branch from Rajar and create a fresh branch testing start. Um while you do that as you can see on the screen there are other things that we have to start looking into drag or skills. So connection is what we currently have. Rajashekar G: rag means like Yeshwanth Reddy Yerraguntla: Correct. Rajashekar G: Salesforce it's another connector so we don't know like Yeshwanth Reddy Yerraguntla: Yes. Rajashekar G: Oracle directly give access like Salesforce or like guit we are doing all the data we need to pull it up and I think okay got Yeshwanth Reddy Yerraguntla: We have to see Manisha Gundapuneedi: Yeah, I think it's Rajashekar G: it okay Yeshwanth Reddy Yerraguntla: Ganesh just one Rajashekar G: okay Yeshwanth Reddy Yerraguntla: second. Rajashekar G: yeah Yeshwanth Reddy Yerraguntla: What is this unifier? What what does it do? How to connect to it? From the sound, it looks like it's a Oracle connector. 01:14:44 Yeshwanth Reddy Yerraguntla: So you have your username, password. It has its own database. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: But until they tell us what it is, we don't know. But what I am saying is um to to to work you Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: don't need to start from scratch already. Manisha Gundapuneedi: I couldn't connect but I Rajashekar G: Yeah, Manisha Gundapuneedi: got Rajashekar G: got Yeshwanth Reddy Yerraguntla: Let me highlight this. Rajashekar G: it. Yeshwanth Reddy Yerraguntla: I'm saying this part you can fetch from enterprise brain and get started but there are other things as per today's discussion we have to do other things also because whatever we are building as enterprise brain before you joined already that is all we were Rajashekar G: Did Yeshwanth Reddy Yerraguntla: building they were building only this yellow path Rajashekar G: I? Manisha Gundapuneedi: Hey, Yeshwanth Reddy Yerraguntla: Addition, concepts, world model, all those things you should continue building so that the way you build it, we can learn from it and make it a template for other Manisha Gundapuneedi: thank you. Yeshwanth Reddy Yerraguntla: connectors. 01:16:06 Rajashekar G: Got Yeshwanth Reddy Yerraguntla: First of all, Rajashekar G: it. Yeshwanth Reddy Yerraguntla: you have to understand that now every connector quote unquote is its own agent. Manisha Gundapuneedi: Let me Yeshwanth Reddy Yerraguntla: Right? This A1, A2, A3, AX, all of these are individual connectors. Manisha Gundapuneedi: record. Yeshwanth Reddy Yerraguntla: A1 is JRA, A2 is Google S, A3 is uh Salesforce, A4 is Unifier. Rajashekar G: Yeah. Yeshwanth Reddy Yerraguntla: And each one of these un each one of these has its Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: own every one of them needs to have all of these things. Okay. So the connector needs to have rag. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: The connector needs to understand certain things about it Rajashekar G: Good. Yeshwanth Reddy Yerraguntla: because because we the the Rajashekar G: Good. Yeshwanth Reddy Yerraguntla: problem with this yellow path is only when user ask something user will get back something. Manisha Gundapuneedi: Basically, we are planning to do it for Yeshwanth Reddy Yerraguntla: Correct. Somehow there is a heartbeat associated with chanaka. every whatever every minute, every 1 hour somehow it has to wake up. 01:17:25 Yeshwanth Reddy Yerraguntla: It has to fetch data from every connector and it has to synthesize that information update its own knowledge base by itself. It has to do it Manisha Gundapuneedi: Thank you. Yeshwanth Reddy Yerraguntla: now. So what I think we should start listing down what are all the immediate things at least the things that we know we have to do Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: right Rajashekar G: So what I'm suggesting is you know first of all like the team has to get clarity on how enterprise brain is working that is one task right away. So in which Yeshwanth Reddy Yerraguntla: when you say the team you mean Manisha's team Rajashekar G: yes so now we are having enterprise brand repo and Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: pyag library all the UI sockets chart interface everything Yeshwanth Reddy Yerraguntla: Yes. Rajashekar G: is in pyog library and we are having a single file Yeshwanth Reddy Yerraguntla: Yes. Rajashekar G: like let's say in a enterprise brain Yeshwanth Reddy Yerraguntla: Correct. Rajashekar G: uh right now like uh today's target we are having already like um to make Salesforce guitars DB three connectors should be in one platform and working well that targeting as of 01:18:44 Yeshwanth Reddy Yerraguntla: Correct. Rajashekar G: now these two are in different branches basic Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: end of the day or tomorrow I'll make sure like everything should be working in like Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: single branch but Salesforce Yeshwanth Reddy Yerraguntla: Got it. Got it. Fall Rajashekar G: back. Yeshwanth Reddy Yerraguntla: back. Rajashekar G: So second thing first of all that is the main clarity the team has to get. Manisha Gundapuneedi: I don't Yeshwanth Reddy Yerraguntla: Yeah. Manisha Gundapuneedi: get Rajashekar G: Immediate Yeshwanth Reddy Yerraguntla: Correct. Rajashekar G: one Yeshwanth Reddy Yerraguntla: prompts one of Rajashekar G: tools and connection Yeshwanth Reddy Yerraguntla: connections basic how to run how to sync and Rajashekar G: both library to enterprise rep. Yeshwanth Reddy Yerraguntla: launch. Yeah, basically you have Manisha team Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: So how to sync and launch is first thing like fetch the repo you do a get clone you do so and Rajashekar G: So Yeshwanth Reddy Yerraguntla: so steps with Manisha Gundapuneedi: Thank Yeshwanth Reddy Yerraguntla: launches Manisha Gundapuneedi: you. Rajashekar G: here question. 01:20:58 Yeshwanth Reddy Yerraguntla: screen that when you say hi, Rajashekar G: Got it. Yeshwanth Reddy Yerraguntla: it responds back saying hi. Yeah. Rajashekar G: 60th version. Yeah, you can directly open enterpriseg.com. Manisha Gundapuneedi: Racist station. Rajashekar G: This is final UI Manisha Gundapuneedi: Okay. Rajashekar G: of samehat. Yeshwanth Reddy Yerraguntla: Okay. Okay. So, screen when you do the right things. Rajashekar G: and dependency. Every time like Manisha Gundapuneedi: That's it. Are Yeshwanth Reddy Yerraguntla: transfer just an Manisha Gundapuneedi: they Yeshwanth Reddy Yerraguntla: ownership. Manisha Gundapuneedi: and then later. Rajashekar G: This is a potential destructive action. Manisha Gundapuneedi: Fore Rajashekar G: All Yeshwanth Reddy Yerraguntla: Okay. Manisha Gundapuneedi: speech. Yeshwanth Reddy Yerraguntla: Dr. Manisha Gundapuneedi: I'm not sure. Yeshwanth Reddy Yerraguntla: GitHub except Rajashekar G: right. Yeshwanth Reddy Yerraguntla: so what you have to do is Rajashekar G: I Yeshwanth Reddy Yerraguntla: enterprise Rajashekar G: know Yeshwanth Reddy Yerraguntla: brain requirements Ideally Rajashekar G: Tom. Yeshwanth Reddy Yerraguntla: this should become just a example Rajashekar G: two versions or something Yeshwanth Reddy Yerraguntla: dire Rajashekar G: link. 01:24:09 Yeshwanth Reddy Yerraguntla: So whenever you make a push you tag it Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: on this is how you private people manage their Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: reposious Rajashekar G: So, pile library updation. Yeshwanth Reddy Yerraguntla: but no pi push Rajashekar G: Got Yeshwanth Reddy Yerraguntla: You will commit. Rajashekar G: it. Yeshwanth Reddy Yerraguntla: You will after committing you'll do a tag get tag and ID on the Rajashekar G: Tags ID on the Yeshwanth Reddy Yerraguntla: Yeah. So every time you come you think something is stable you'll tag it. Manisha Gundapuneedi: All Yeshwanth Reddy Yerraguntla: That's how you will update it here also. Manisha Gundapuneedi: right. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: So this I'm now moving it to who is this Manisha Gundapuneedi: All right. Rajashekar G: The Yeshwanth Reddy Yerraguntla: transfer specify design Rajashekar G: bomb. Yeshwanth Reddy Yerraguntla: labs. Rajashekar G: The bomb. Yeshwanth Reddy Yerraguntla: Uh I will do one thing before I do this. Manisha Gundapuneedi: Perfect. Yeshwanth Reddy Yerraguntla: I will settings uh collaborators slow. I will add a Rajashekar G: 18 individual address people. Yeshwanth Reddy Yerraguntla: I know I know like you can ask Raia to give access to everyone. 01:26:02 Rajashekar G: Sure. Manisha Gundapuneedi: Okay. Yeshwanth Reddy Yerraguntla: Disable transfer insufficient permission. Yeah. Rajashekar G: Choose a number. Manisha Gundapuneedi: Specify an organization lecture. Yeshwanth Reddy Yerraguntla: Okay, cool. Rajashekar G: Yeah. Yeshwanth Reddy Yerraguntla: Collaborator seats to create private repositories Manisha Gundapuneedi: That may collaborate Yeshwanth Reddy Yerraguntla: on Manisha Gundapuneedi: transfer. Yeshwanth Reddy Yerraguntla: change visibility change to public. Do not does not have enough collaborator seats. They would Manisha Gundapuneedi: already transfer Yeshwanth Reddy Yerraguntla: income Manisha Gundapuneedi: into Rajashekar G: Wow. Yeshwanth Reddy Yerraguntla: currently public. Rajashekar G: But still time. Yeshwanth Reddy Yerraguntla: Monday. Rajashekar G: Okay. Manisha Gundapuneedi: Yeah. Yeshwanth Reddy Yerraguntla: Monday. Manisha Gundapuneedi: Thank you. Yeshwanth Reddy Yerraguntla: I think I know what is happening. Manisha Gundapuneedi: I'm required data Yeshwanth Reddy Yerraguntla: He's outside the organization. Manisha Gundapuneedi: from Yeshwanth Reddy Yerraguntla: So now it should work. Sorry. until the new owner approves. I want to transfer this Rajashekar G: Goodbye. Manisha Gundapuneedi: That's Yeshwanth Reddy Yerraguntla: We still need to list down whether what things we have to do. Manisha Gundapuneedi: something 01:29:28 Yeshwanth Reddy Yerraguntla: So one is uh uh Manisha squad needs to sync with enterprise Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: group. Manisha Gundapuneedi: just Yeshwanth Reddy Yerraguntla: Next is you are ensuring All connectors are available, right? Manisha Gundapuneedi: Yeah. Rajashekar G: M Now I'll share the branch. I'll share the env any connectors available in environment is they can easily able to understand Yeshwanth Reddy Yerraguntla: Yeah. Yeah. Yeah. That's what I meant. Rajashekar G: what's happening Yeshwanth Reddy Yerraguntla: Next task is um Manisha's Rajashekar G: there. Manisha Gundapuneedi: Get back. Yeshwanth Reddy Yerraguntla: squad to give uh KT on how Manisha Gundapuneedi: What's Yeshwanth Reddy Yerraguntla: um Unifier works. Oracle unifier. Rajashekar G: Yes. Yeshwanth Reddy Yerraguntla: How is Oracle work? Manisha Gundapuneedi: happen? Yeshwanth Reddy Yerraguntla: KT you have to understand how it works. Rajashekar G: No. Ganesh Danuri: Once she's done with it, we'll take 80 from Manisha Gundapuneedi: Okay. Ganesh Danuri: her. Rajashekar G: One Yeshwanth Reddy Yerraguntla: So while that is happening um yeah you guys can start on this Rajashekar G: second. Yeshwanth Reddy Yerraguntla: parallel track game. 01:31:13 Yeshwanth Reddy Yerraguntla: Yeah, essentially once you are clear on step Manisha Gundapuneedi: Come here. Yeshwanth Reddy Yerraguntla: one, uh you have to do a bunch of things. Uh rag uh uh DB build rag DB for uh trained patterns. Training patterns skills for general knowledge, general patterns. Uh building a world model. Next chess. um snapshot and escalation triggers. Rajashekar G: There Yeshwanth Reddy Yerraguntla: Sorry. Manisha Gundapuneedi: Stop Yeshwanth Reddy Yerraguntla: Snapshot entry. Rajashekar G: go Yeshwanth Reddy Yerraguntla: You have to basically teach all of these things. Manisha Gundapuneedi: it. Yeshwanth Reddy Yerraguntla: Uh escalation trigger just Manisha Gundapuneedi: Welcome Rajashekar G: world model platilla. Manisha Gundapuneedi: to Yeshwanth Reddy Yerraguntla: world model is what is expected out of unifier. Oracle unifi what is being expected and what are the things it is working Manisha Gundapuneedi: happy. Rajashekar G: H. Yeshwanth Reddy Yerraguntla: with what are its concepts so concepts tasks acceptance criteria Rajashekar G: Mhm. Yeshwanth Reddy Yerraguntla: epic story users projects we how are they interrelated to each other and H no clarity is to know how so what is it what is it even dealing with snapshot on 01:33:27 Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: the other hand is on a specific day what is the state of Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: Jira for every project for every epic for every task what is the progress all that is snapshot multiple snapshots are needed for the name if I call it s1 and s_ub_2 two somehow if it can compare these two only then it Manisha Gundapuneedi: You can see Yeshwanth Reddy Yerraguntla: will know something progressed or something didn't progress so Manisha Gundapuneedi: Oh. Yeshwanth Reddy Yerraguntla: you can ask me why do I need to compare and see why something progressed or not that is because when something does not progress you know that you have to escalate it escalation is how you make Manisha Gundapuneedi: Oops. Yeshwanth Reddy Yerraguntla: the whole agent proactive Okay, after that. Rajashekar G: But I have one concern here. So majorly what we felt while working with Salesforce Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: Do we have anything like the Yeshwanth Reddy Yerraguntla: I think that that needs to be fetched to first thing from I don't Anita Rajashekar G: structure 01:34:55 Yeshwanth Reddy Yerraguntla: job or whoever it is the fetch Oracle because the Tata people are giving their own Rajashekar G: and Yeshwanth Reddy Yerraguntla: connector. Rajashekar G: not talking about connectatformatform what purpose it then we'll get to this connector in that gives much better clarity Uncle Yeshwanth Reddy Yerraguntla: K. That is part of the K in my head. Rajashekar G: already existing. Manisha Gundapuneedi: Foreign speech. Foreign speech. Foreign Yeshwanth Reddy Yerraguntla: only as soon as Manisha Gundapuneedi: speech. Rajashekar G: Okay, Yeshwanth Reddy Yerraguntla: possible. Rajashekar G: got it. Yeshwanth Reddy Yerraguntla: You think whatever you want to call it. Cadence is basically the heartbeat of Rajashekar G: The cadences client name prospect Manisha Gundapuneedi: Um, so use Rajashekar G: name Yeshwanth Reddy Yerraguntla: English term cadence and uh that is what I meant by Rajashekar G: English. Yeshwanth Reddy Yerraguntla: cadence. Rajashekar G: Okay, got it. Yeshwanth Reddy Yerraguntla: Cadence is regular walk followup. Rajashekar G: Got it. Got Yeshwanth Reddy Yerraguntla: Yeah, Rajashekar G: it. Yeshwanth Reddy Yerraguntla: this the time duration between S1 and S2 is the cadence or delta after 6 hours what is the current snapshot? 01:36:41 Yeshwanth Reddy Yerraguntla: So your cadence is 6 hours Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: once the DB connector is available. Sorry. Four is to basically integrate EV into Manisha Gundapuneedi: Let's Yeshwanth Reddy Yerraguntla: sorry integrate Oracle unifier into enterprise basic connector Manisha Gundapuneedi: just Yeshwanth Reddy Yerraguntla: test these all will come into picture Manisha Gundapuneedi: take Yeshwanth Reddy Yerraguntla: now 5.1 5.2 2 5.3 5.4 dot dot dot but we are doing this because we want to strengthen enterprise brain Manisha Gundapuneedi: Sorry, Rajashekar G: Got Manisha Gundapuneedi: guys. Yeshwanth Reddy Yerraguntla: next we see six finally six is not Rajashekar G: it. Yeshwanth Reddy Yerraguntla: six in the sense it's parallel testing framework Manisha Gundapuneedi: as in manual. Yeshwanth Reddy Yerraguntla: Yeah. Manual. Manisha Gundapuneedi: So we need to see how we can automate Yeshwanth Reddy Yerraguntla: Yeah. Manisha Gundapuneedi: using it ease things up and also we will not miss any Yeshwanth Reddy Yerraguntla: Yes. Manisha Gundapuneedi: use. Yeshwanth Reddy Yerraguntla: This should happen in parallel. So this is all as far as uh Manisha's team is concerned Rajashekar G: Yeah. 01:38:42 Yeshwanth Reddy Yerraguntla: but Raja's team is mainly looking at sorry Rajashekar G: Second. Yeshwanth Reddy Yerraguntla: discuss Manisha Gundapuneedi: Time and Rajashekar G: Third point. Core changes Manisha Gundapuneedi: press. Rajashekar G: orchestration. Yeshwanth Reddy Yerraguntla: correct correct correct correct So Manisha team is focused on this blue boundary. this blue boundary whereas I have Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: your team will focus on weirdly Manisha Gundapuneedi: I Yeshwanth Reddy Yerraguntla: this this Manisha Gundapuneedi: guess Yeshwanth Reddy Yerraguntla: boundary a green if I isolate it what I'm saying Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: is you have your core chanaka a z a z is responsible for doing all sorts of communication with every Rajashekar G: H. Manisha Gundapuneedi: you Yeshwanth Reddy Yerraguntla: other agent A1, Rajashekar G: Got Yeshwanth Reddy Yerraguntla: A2, A3. Rajashekar G: it. Yeshwanth Reddy Yerraguntla: So the reactive proactive first thing is we have to make sure it is proactive. We have to like facility Rajashekar G: Auto syncing basically. Yeshwanth Reddy Yerraguntla: syncing in one sense. Yeah, it is autos syncing but it's not only autos Manisha Gundapuneedi: Oh, 01:40:05 Yeshwanth Reddy Yerraguntla: syncing the user happy user ask Manisha Gundapuneedi: what is it? Yeshwanth Reddy Yerraguntla: something user ask something right that is the happy we are Rajashekar G: Hey. Yeshwanth Reddy Yerraguntla: saying a z needs to have its own uh concept of uh state of diwame. Rajashekar G: Thank you. Yeshwanth Reddy Yerraguntla: So a z has its own concept of a knowledge base, its own concept of a rag, its own uh cadence. every so many hours. So I have it has to ask every agent by itself. Hey, what is the update of Jira? What is the update of Salesforce? It rag uh rag and vict rag is mainly necessary for understanding user preferences, user behavior, uh some organization preferences, connector level cadence, right? Connect connector cadence. Maybe I need every 1 hour males I need every 1 minute right what so this this Rajashekar G: All right. Yeshwanth Reddy Yerraguntla: is sort of acting as a configuration layer for a z knowledge base on the other hand is saying what are the active projects right what are the uh user preferences user roles roles and 01:42:05 Manisha Gundapuneedi: It's Yeshwanth Reddy Yerraguntla: responsibilities point. Manisha Gundapuneedi: very Yeshwanth Reddy Yerraguntla: Um Navinana was saying right at madam responsibility is to ensure customer satisfaction is there to ensure uh sales pipeline is healthy to ensure lead management is healthy those are our responsibilities based on that even inside Jira when something's happen some kind of relevance need to go to her that sort of Manisha Gundapuneedi: That's Yeshwanth Reddy Yerraguntla: indicating customer will not be happy eventually. Right? Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: So those kinds of information need to sit in this rag. Manisha Gundapuneedi: good. Yeshwanth Reddy Yerraguntla: KB on the other hand is saying what is the state of Dwami itself. There is a project called Rajashekar G: Mhm. Yeshwanth Reddy Yerraguntla: Tatastel Salesforce pipeline. Nobody filled it. Rajashekar G: Mhm. Yeshwanth Reddy Yerraguntla: So I have to escalate it to someone uh something else. Here are the meeting all the meeting links that are in Tata Steel uh some other project infinity is one right something else is happening here. These are all the relevant information. 01:43:27 Yeshwanth Reddy Yerraguntla: So this is like a real time status of what is happening there. So that uh most of the questions can be directly answered from here. Rajashekar G: Good Manisha Gundapuneedi: Yeah. Take care Yeshwanth Reddy Yerraguntla: Next thing that you have to build is the uh cadence Rajashekar G: idea. Manisha Gundapuneedi: of Yeshwanth Reddy Yerraguntla: engine that you know you have to every 10 seconds call this agent by saying Manisha Gundapuneedi: Daniel. Yeshwanth Reddy Yerraguntla: hey it's time is 12:00 do you want to do something then agent might decide okay it's 12:00 I let me let me fetch something from Jira or let me fetch something from image. Rajashekar G: Got Yeshwanth Reddy Yerraguntla: It will it will fetch it will update its knowledge base. Rajashekar G: it. Yeshwanth Reddy Yerraguntla: It will also decide the last thing that you have to build which is the uh notification engine or whatever you want to call it. Notification not exactly as mobile notification. It can mean um I will decide I am going to put these few things in the dashboard for so and so user. 01:44:39 Yeshwanth Reddy Yerraguntla: every user dashboard create right. So after every hard bit it might decide okay u okay this Rajashekar G: Hello. Yeshwanth Reddy Yerraguntla: is a e particular task is uh having a bug. So I have to notify pratimagaru that something is going to go wrong in her own language. I have to notify prabakar in his own language with the right evidence. Rajashekar G: I'm Yeshwanth Reddy Yerraguntla: So all that it has to now start deciding sometimes very few times it will if it is really critical it will Rajashekar G: sorry. Yeshwanth Reddy Yerraguntla: send a WhatsApp message WhatsApp andagani we didn't decide it yet but it can send those notifications or it might simply say I'll send an email these while connector is doing a lot of heavy lifting chanaka itself is doing even more heavy work. We have to start planning how to do it, how to uh put them in the form of what are the necessary tasks and how to distribute the work. Rajashekar G: Got You don't Yeshwanth Reddy Yerraguntla: So what I'll suggest 01:45:53 Rajashekar G: know Yeshwanth Reddy Yerraguntla: is what are the things that can immediately be demonstrated like feeling should like at the same time I don't know Rajashekar G: any complexity like bits and pieces. Everything is a small thing only. It's all bits and pieces. It's good. But everything Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: No, Yeshwanth Reddy Yerraguntla: Open. Rajashekar G: we can make Okay. Yeshwanth Reddy Yerraguntla: How can we improvise and bits and pieces implement discussions? Maybe you can come up with the right plan and who will demonstrate it. Rajashekar G: Right. Shut Yeshwanth Reddy Yerraguntla: anything because Rajashekar G: up. Yeshwanth Reddy Yerraguntla: all of them are equal important. Rajashekar G: Oh no no no like I'll involve everyone like two teams individually meeting and we'll involve Prabhakaru also. Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: So by keeping everyone there how fast they need. So also like guide us Yeshwanth Reddy Yerraguntla: Yeah. Yeah. Rajashekar G: better. Yeshwanth Reddy Yerraguntla: Yeah. Yeah. Yeah. Business parangologist they will have their own priorities. 01:48:36 Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: So for example notification even though there is no rag or knowledge base it's still fine to some extent it can do something Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: I will share Rajashekar G: Yeah. This is third part, right? Complete. Yeshwanth Reddy Yerraguntla: Third part on the third part. You're right. Rajashekar G: The first two parts are like Yeshwanth Reddy Yerraguntla: Correct. Rajashekar G: the third is how it works like what should All Yeshwanth Reddy Yerraguntla: Correct. Yeah. Correct. Rajashekar G: right. Manisha Gundapuneedi: So, Yeshwanth Reddy Yerraguntla: level. Yeah. Yeah. Every Manisha Gundapuneedi: so Yeshwanth Reddy Yerraguntla: connector Rajashekar G: Multiple Yeshwanth Reddy Yerraguntla: is what is Manisha Gundapuneedi: I can do Yeshwanth Reddy Yerraguntla: that? Manisha Gundapuneedi: connector. Yeshwanth Reddy Yerraguntla: RA skill set world Manisha Gundapuneedi: Keep Yeshwanth Reddy Yerraguntla: model and trained patterns patterns patterns. Okay. Manisha Gundapuneedi: global system Rajashekar G: Escalation Yeshwanth Reddy Yerraguntla: How do Manisha Gundapuneedi: as in overall enterprise brain level. 01:51:06 Manisha Gundapuneedi: Overall enterprise brain level escalation trigger. Rajashekar G: trigger. First of all, so something happens in this database. Let's say Jira database Manisha Gundapuneedi: RIP Rajashekar G: something. Yeshwanth Reddy Yerraguntla: It can be as simple as asking Oracle. Do whatever you want. Rajashekar G: Let's say multiple minutes that should fetch from the rag at 10 what we need to trigger whether it's a notification either it's a UI dashboard whatever Yeshwanth Reddy Yerraguntla: Aranga. Rajashekar G: am I clear I think just basic level high level easy but we need In depth law there will be Manisha Gundapuneedi: invest. Rajashekar G: things. This is okay like we'll do the like task breakdown end of the day in Yeshwanth Reddy Yerraguntla: H. Rajashekar G: my mind how I'm seeing it as so whenever we jump to the execution Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: previousity for this one while planning itself I want like good how management is expecting is the process should be every story should have maximum 40 hours only and a story individual Manisha Gundapuneedi: That's Rajashekar G: owners any task shouldn't be exceed more 01:53:44 Manisha Gundapuneedi: good. Yeshwanth Reddy Yerraguntla: H. Rajashekar G: than everything whenever if you go to some other prospect and demonstr I think we'll be much more crystal clear with the efforts we we are going to Yeshwanth Reddy Yerraguntla: H. Rajashekar G: put. So that's why that's where also will help us much better story breakdown and task story Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: breakdown then we are good enough on the plan we will try to Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: work. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: Okay. Manisha Gundapuneedi: Okay. Rajashekar G: And one more thing regarding the simulator. So genius just has Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: shown an Excel sheet Manisha Gundapuneedi: with all the arch types. So different types of people, Rajashekar G: connection. Ella able to Yeshwanth Reddy Yerraguntla: Okay. Manisha Gundapuneedi: different intentions, sorry intent of uh so different forms of data, Rajashekar G: remove. Manisha Gundapuneedi: how do we represent them? End of the meeting. Sorry. Rajashekar G: I want to correct something like questions. It's like inputs. I say input he's a CXO. 01:55:33 Rajashekar G: He's a fast decision maker. He likes Manisha Gundapuneedi: question. Rajashekar G: uh Manisha Gundapuneedi: Here are few things that you need to take care of. I'm trying to Rajashekar G: like this kind of person is he likes this kind of visualizations. is a fast decision maker based on our framework JSON trigger and we need to tell the scenario also that is major thing like Yeshwanth Reddy Yerraguntla: H. Rajashekar G: you are already delayed by 10 days when we run it from the framework it should show the dashboard a dashboard visually appealing everything should be there similar so what I'm thinking is there are two ways one is uh Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: okay agent agent knowledge we will be giving a input likewise and it will be generating a plotly chart or something in the UI and second scenario is directly conversing with chart Yeshwanth Reddy Yerraguntla: Mhm. Rajashekar G: GPT Okay, Yeshwanth Reddy Yerraguntla: That is how we should do it because we are still exploring. Rajashekar G: got it. Sure Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: enough, Manisha Gundapuneedi: It's done. 01:57:15 Rajashekar G: we'll put it in. We'll be directly putting in the prompt. Manisha Gundapuneedi: Okay. Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: Yeah. Anything else in that Manisha Gundapuneedi: No, Rajashekar G: topic? Manisha Gundapuneedi: first we need to check from the priority. on because that is one of the action items Rajashekar G: Okay. Manisha Gundapuneedi: from the meeting and we both Rajashekar G: That's it. Yeshwanth Reddy Yerraguntla: Uh prompt drop. It's okay. Nightly. Um. So test evaluations force list tables just uh tables uh table I'm asking Abhilash Adunuri: Okay. Yeshwanth Reddy Yerraguntla: you to simulate Salesforce first thing_ system. So what you should do is ask this question with different system prompts old uh ask this question to different models 2.5 flash 2.5 flash 3 pro 3.1 how the Abhilash Adunuri: Yeah. Okay. Yeshwanth Reddy Yerraguntla: models are reacting to old new table ask it like Abhilash Adunuri: Okay. Yeshwanth Reddy Yerraguntla: same question 10 times 10 times 7 * 2.5 table information failure rate is 70%. Abhilash Adunuri: Yes. Yeshwanth Reddy Yerraguntla: 3 flash issue Abhilash Adunuri: Yeah, I assume. Yeshwanth Reddy Yerraguntla: probably model issue. Abhilash Adunuri: Yeah, but we'll try that. Yeshwanth Reddy Yerraguntla: Yeah. matrix. Abhilash Adunuri: Yeah, we'll do it. Yeshwanth Reddy Yerraguntla: Yeah. Abhilash Adunuri: Oh, okay. Yeshwanth Reddy Yerraguntla: Yeah. Clarity. No. Chip. Chip. Abhilash Adunuri: Fine. Okay. Yeshwanth Reddy Yerraguntla: Chip. Abhilash Adunuri: So, validated tool. Yeshwanth Reddy Yerraguntla: God, I can understand. Sorry. Abhilash Adunuri: I know. Yeshwanth Reddy Yerraguntla: Correct. Abhilash Adunuri: Better if Yeshwanth Reddy Yerraguntla: Mhm. Abhilash Adunuri: I should not do that. You are capable to do that. try on quy but we are not we not Yeshwanth Reddy Yerraguntla: Okay. Abhilash Adunuri: validating and quy Yeshwanth Reddy Yerraguntla: Oh, okay. Done. Abhilash Adunuri: valid Yeshwanth Reddy Yerraguntla: I see what you mean. Abhilash Adunuri: Now on the internet Yeshwanth Reddy Yerraguntla: Always lightweight. Abhilash Adunuri: Are we chasing matrix playlist? Yeshwanth Reddy Yerraguntla: Okay. Sorry about Abhilash Adunuri: Thanks. Transcription ended after 02:03:35 This editable transcript was computer generated and might contain errors. People can also change the text after it was created.