Feb 9, 2026 Enterprise Brain Design Discussion - Transcript 00:00:00 Rajashekar G: So enterprise brain the concept remains same as a platform we are going to create a platform which will capture different aspects of enterprise. So here um for now so the concept the actual concept of enterprise like so in every department if you're talking about enterprise grade companies where they will be having a multiple departments and each department is dealing with the multiple data sources to complete their fulfill their activities or jobs. going to talk to a single database single source of truth or it can be connected to the multiple databases to view that but it's not giving a view especially when you go to the V kind of an interfaces where there are very vast systems the database and all so it's not cross origin is not going to happen there database uh we are able to connect to the multiple database to tell a common story is not coming up so we are we are seeing a lot more potential from a business and need perspective uh where the one person should know about what happening in different department especially we are targeting for sees people that the entire the the platform is going to we are focusing primarily on the audience who are sitting at a se level. 00:01:25 Rajashekar G: So why right now right these people are very strategic people and they know how to drive the innovation and they are the people who are setting up the vision but these people are always dependent on the other leadership team to provide the right information to them. So there is always they feel like they're dependent on someone they're not like a free flow because similarly right because uh as an individual contributor you know what you are doing what you're delivering and you feel like you are more comfortable in doing the things as a manager I am dependent on you people to give me a real status if something status is not clearly reflecting that so the project may go high five the same way the as a three minutes is a level of where the issue maker division is set up those people are dependent on the information to receive now right now that is fulfilled by the next level of leadership team and they are dependent on the next level of leadership team. So it's like everything because they're not going to go and work on individual activity right because we need that's the reason the people say that we need a strong inline next inline people to tell you the true facts of what happening there so the enterprise B is going to fill that gap because instead of because the qu questions right maybe today PMA joined the meeting and tell that she said that I need this answers for these questions then next time when we come to this meeting we always come with those questions answers we already prepared those answers for those questions but 00:02:46 Rajashekar G: she will ask another question. So every time we don't have any readym made answer to answer the question of the CXOs because we need to prep right because we are also depend on the next level of people to use the data. Now the enterprise brain is like a very dynamic in its own. Oh, now it's connect to the multiple departments, right? For example, if I want to run a manufacturing industry, there is a planning department, there is an procurement department, there is an head of the the unit uh department which runs the production, there is a quality department. In order to get an understanding of what whatever I produced in the yesterday and what is the quality output and what is an error rate and what is the defect rate if you want to know all the people has to be sitting in a common room and they have to come up with their own reports. Then as a strategy person I need to consume those report to give that. Now assume that now enterprise bane is there which is talking to the quality data which is talking to the procurement and trying to uh analyze those data and giving an insights on top of that. 00:03:45 Rajashekar G: How much it is helpful for the decision maker to take a quick decisions. So that's what we are building here. It's not about a single platform single product because it can be any department any data source. It can be snowflake red shift Oracle SQL doesn't matter where your data lip may know how to connect to the data and how to right now our focus is not on like cross origin and comparing those things at least if you connect to one database you ask your question in a natural language system can interpret in a better way right now our analytics are also divided into four categories one is like a descriptive descriptive analytics nothing but you have the data you're just showing the facts That is a descriptor. That's what we are showing right now. Based on the data, we are showing some sort of insights to it. Your production is this thing or your your fill rate is this much and all that is like true facts you are putting in a different way for the users to catch. 00:04:40 Rajashekar G: Now there is an diagnostic layer or inside layer we can say. So it's more about what we what is that is going to tell looking into your data a true facts it is going to tell last month your fill rate was 90% now it was dropped by 2% is this 88 so it's giving some insight to you further elaborating you to take a calls in a quicker and faster way then we have a diagnostic layer now you're trying to understand why might percent rate 2% drop happen so someone is trying to understand why the reason behind why there so for that the system need to go to back backtrack all the lags activities and it's going to tell that this is the final root cause of why there is a dropage 2% in some line there is an equipment failure so you're not able to produce that or there is a planning issue and um the manpower shortage so because of end reason the system is not able to achieve the what you are doing the last time so that is a analysis now now it's a mitigation like predictive analytics is coming into the picture prescriptive on prediction is more about now I'm in this state now how can I make it better in the next week or next month or this thing so system is going to tell you predictions and recommendation what you can take for example now operate the second line in a second shift one more shift so that your productivity or fill rate will go high or you need to procure a because last time material shortage now you need to find an 00:06:07 Rajashekar G: alternative supplier to produce that material to increase that productivity or you need to make sure that manpower work for it to one more additional hour for the one week to make sure that film. So these are the different recommendations and that is also the now the predictions comes if I take this recommendation what is my probability of fitting that. So like that we we have a multiple layers we are building right now from day one right now we are beginning with the descriptive layer it depends upon how we want to do with every client to client we are building a platform now you can able to connect to the data source you can ask any questions I can able to show a descriptive analytics to you mean two facts to you now if I have a more capability of doing some sort of reasoning and insights driven we can I can able to give an insights to you now if you still I can mature my product in telling that I can able to tell the root cause analysis that's where you will go to the diagnostic layer. 00:06:59 Rajashekar G: So even though the application is same based on the maturity of the product or expectations we can increase the layers that we want to build on top of that. So this is the enterprise brain concept to tell that so it serves all the needs of the uh people because you can tell that that can be used by the manager but it's like you can limit your arbback there as a manager what data points you can access as a department head what data points that you can access always it's like restricted access permissions and arbback defined at a data layer based on that if you query that you can able to system can able to answer that a lot more governance lot more security will come into the picture but at a broader level you can consider it as a a a assistant where which knows everything about the company you're simply asking to get an answers and it's like there is no delay in answering your question it's like it's more about how you are vision driving we are telling that personalized assistant and personaliz experience right based on what kind of a questions what kind of vision drives our uh enterprise brain also matured enough to personalize it to that particular visionary or a stakeholder that's the vision that we want to take it the product. 00:08:10 Rajashekar G: Now why we are gathering here is like and it we are seeing a lot more potential of how the potential of this product being to a next level and useful. Now our challenge is that how to present this product to the external fold. We don't want to go in a regular way of presenting it and one the PCs are ready the enterprise B is able to connect to the snowflake or kill DB at this point of time and able to answer where the inside descriptive layer is done but the presentation layer is missing the same thing we don't want to go with the regular RGB way of presenting that information so what is the right way to come up with to present the enterprise bane uh kind of a thing because we are bored with seeing that uh prompt based communication collaboration based charg interface and all that is there any better pattern that we can able to bring that's what the the context of this workshop just sharing the ideas and the vision of the product I just elaborated at a very high level once is here she might give a lot more context to what we want to achieve so this is where we are in uh and yogi is part of the cause but we didn't made any problem same problem Because enterprise is the same thing. 00:09:24 Rajashekar G: Visual defect is like a part of the enterprise B. It can be one more uh platform or application coming into sitting on the platform. It's just focusing on defect identification in a manufacturing unit. So like that enterprise brain is like a base layer which can accommodate all types of needs that the enterprise needs and everything. So that's the reason we don't want a regular dashboard there. We are seeing that how can we enhance the experience just to give a feel like it's an AF first product. One is from an not only from an ideation point of view a strong core idea but until other we don't have a strong visual and we want to go with an experience-driven approach here. So that's where we are struggling because even though we are putting individual thoughts there we are not able to crack. So I thought that if the bigger team is involved in it, we can able to discuss, brainstorm and at least come to some common ground. What would be the right thing that we can able to achieve? 00:10:22 Rajashekar G: Uh would you like to add anything is a DM for this project. So uh one thing is we need to be able to support all modes of input and output. Basically we will be able to enable the voice based inputs also. Currently we only have text based input. We may also give voice based input and uh it may also be able to there may additional tools built around it going forward. As of now it is only a prompt and a response. Even the response is also like text uh tables and visuals which is drill down kind of the graphs and charts which can be further drilled down to get the right data. It can also be possible that considering some of the use cases we may also have to have some kind of comparison utilities like last week I have produced version one of a certain document now I have version three now what has changed in between these two uh from that from that what are the business insights that I can derive that is one and next uh for each user it should be able to maintain the context similar to how charg maintain you have different context windows when you switch to that context window per user it will find out that okay this is what it is. 00:11:36 Rajashekar G: It also learns um some business language and uh when it is not clear on certain terminology it may also further ask you for clarifications that is another part. So and we are also providing we wanted to provide a feedback mechanism wherein if you think that the response given is irrelevant or incomplete you may give a reaction to it and that would ask you what are the reasons for it. I think we have implemented the same thing in IR similar to that. They implement everything but everything looks similar to from air and was uh even we did in the one of the project right last time the feature last we introduced any AI feature collaboration collaboration no I think I was trying to um what was I the CSS load I was trying a experiment where I could in CSS we implemented it root cause analysis is again we try to do that so from there Converse every interface looks similar because we are not able to think beyond because that is a great pattern someone introduced. It's feel like it's convenience or it looks like it's natural interaction. 00:12:49 Rajashekar G: We don't want to deviate from there but we need to see an alternative pattern because again the problem with that attraction is the context is missing every time I'm coming to the window with a fresh context of to begin with the journey. So now assume that the context the trigger has to move from the context because I don't know whether I g a white canvas you can start at any corner of the page to begin with instead of going a traditional approach of you start your chart and corners something assume that even though you give a white canvas the people are struggle to start their journey you need to have a little more guidance to them where to begin and all that so for this kind of a platform how can we give that guidance And we don't want to go with a chart GPT kind of an experience where you have created one your GP right there is a different GPT there when you go inside again interface will look like. So now assume that we should not uh load the user with multiple GPT interfaces. 00:13:47 Rajashekar G: So they don't know which GP to invoke. So assume that it's a simple an interface for me. I can ask whatever I want. The internal logic should define which agent to route. Maybe that same query need to be routed to three different agents. It will route to three different agent and three agents will give a different kind of a responses. Our front end layer or one more agent has to consolidate those responses and give a single one. So we are not limiting to an agent to agent communication. This is a multi- aent communication because the same question can pass to different agent to analyze for different different things but the we need to summarize that responses and we need to give a common story. So like that there are multiple strong use cases are there here. So now just we need to think about what is the pattern that we can able to think as even in the optin we move that thing and we put it as a history on the top just to break the monotony of being the same layout. 00:14:49 Rajashekar G: What did we do? How do we know the chart history will be here right in chbd we remove this and we put it as an option as history that's it rest interface will look similar only so our brain stopped here we are not able to think beyond and one more pattern right same you ask for a query and it's generate a graph here at the end right so it's a graph but what are the other things that we can do on top of it it's coming in a bound box. Now assume that even in the table is big having a multiple columns you are limiting your thinking to give a common grid with the pagenation component at the bottom. Why we have to limit to those patterns why can't we? Because at least the query engine can be a common thing but the response has to take its own form like somewhat I feel like better with the cloud and all this interfaces where the document is generated right it switch to its context view it won't leave it to the provided there document there it will take to a different view all together create a different document and give an interface like that we need to think here based on the responses if the user needs a big canvas we need to give a big canvas to see that response because table and all this. 00:16:05 Rajashekar G: So these kind of a patterns we need to look into typically what will be there the input is only three or three or four types there is an audio video text uh only this three other three ways of inputs right a file upload or you can put a file as an upload but output can be any number of things but we again we are putting into the four ways only either it can be a graph it can be a table it can be a document or it can be an image but when you're showing a document but the document has its own template and structure in In case of optin we define the structure right root cause analysis how to show that. So like that every query has its own structure to define but again optin what we did we put it in the same chart window and we restricted the view. For example the chart the network diagram came into the picture. Why the network diagram has to be a thumbnail by when you click on it then when user has to come implicitly and tell that expand to view that kind of a view. 00:16:57 Rajashekar G: Why? And the response come by default why can't be that view to the user because we know that user is interested to view that response right then why we have to put in a small window which is not readable to the user. So think about those patterns because when the intent is very clear what the user is going to do in the next step our patterns has to adapt to that instead of going with a regular thing because right now the tech visibility has gone because in internal product they just challenging the design team you give us a pattern we can able to accommodate but as a design we are not able to think beyond due to control factors because we don't you people may say that we don't have a time to explore but we need to see alternative but not the same patterns Changing is okay. But these are the used the pattern. So for relatability point of view for user by seeing this one itself they easily understand. If you break the pattern then they will struggle to understand what it is. 00:17:55 Rajashekar G: But this pattern is not previously established. Someone has introduced this the charg pattern. No no I'm not talking about that. So you're saying a table and graphs you're waiting that patterns no one tried when me right you know that I want to explore a table so for me you are making it two click to open the table in a full view because I ask you a question what show me the performance of uh the projects that worked in the last 6 months it's it's going to give a table to me or it's going to give a graph to me your right now interaction pattern is limited to a thumbnail kind of a view where I can't even read So you're expecting me to click it again. For me, from an end user perspect, I feel like I ask the why can't you show me that up front to me? You're you're you're honestly making me to click one more thing. Right. So those are the things those are the experiences that people won't realize that I'm doing that. 00:18:51 Rajashekar G: So to get the clarity I I agree with your thumbnail. It can't be show anything after that. One more click you're introducing instead of that showcasing up front is what you are suggesting that is one of the example table will be table graph will be graph. Yeah because table will be table graph will be a graph but if you feel like that can be a very massive table where you can give an immersive experience there yes we can able to give because you feel like there is a root cause analysis you want to tell in a narrative way yes we can able to do there because I'm not limiting the pattern here because table is one because the same data table can be represented in a graph view right but we are changing that perspective why it has to be always a based on the context based on the industry based on the persona that system will take a shape of that thing. Right now I'm not even telling that what is the alternative for table but at least the basic patterns what we are showing can be enriched okay 00:20:00 Vinuthna Srinivas: Can I show you this? Rajashekar G: so this is the enterprise brand so these are all the history So if I come and ask a question here. So don't don't show don't show influence. So now the a team has shown us few capabilities to us. What they shown is like the page need not to be created before. For example, the previously right, we used to create a pages for example a booking details page right now. So in the right now in the current implementation that there should be a booking details page user there is a listing page there is the details page like that every page is designed or implemented by engineering team to make it available to the end user. Now the engineering team has shown us the capab team has shown us the capabilities. The page need not to be created before we can create a page on the fly because assume that it it has an ability to create a different templates and create a page on dynamic ways and it can expose to the user. 00:21:14 Rajashekar G: So based on the information that needs to be presented it form a new pattern new pattern always but we need to assume that those patterns are the templates we are giving. That is what I was asking. So, so like that if this kind of a data need to be represented we'll tell that these are all the template that you can explore for the table format the system can have an intelligence to pick that for this kind of an output I need to go and fetch that pattern template and I can use the second thing they told right uh because right now the charg whatever you ask it will give a simple graph or a table or a text or image but interaction on top of that is not possible they showed us a pattern where it's not about generate the graph it will allow you to build on further it's an interactive graph is what you can able to build the third one they told us it's not about a single visualization I can able to generate a dashboard to you if the whatever you ask if that need to be explained in a story the multiple graphs with the multiple information they can able to bring the dashboard like Tableau dashboards if you guys explore the Tableau dashboards where it has a multiple story where it has a common filter on the top if you alter the 00:22:22 Rajashekar G: filter everything on changes right like our current dashboards only like how our current dashboard personal dashboards we are building on the fly they can able to create a dashboards now they're showcasing all the capabilities what they can able to do so but our solution whatever they built is not reflecting them it's it's just showing as a PC level as an individual concept it's life now how can you make that as a better story when you're building the enterprise way that's another thing that we have to work on literally they I'm impressed with the the dashboard that is generated on dynamic basis. Previously they have to relay the process is very tedious there. They have to uh set the expectations for a business analyst. So business analyst will talk to a Tableau guy to create that experience on a tableau and the dashboard need to be exposed to the enterprise grade uh people to see what happening on that. Now if that kind of a capability is just a one query away see that how much potential we are making there. So now all these things how can we make it unstitch in enterprise B is what we have to do that demo I have only this is like table plus part so same pie chart is in a 10 format here so this one this is a long prompt I'm giving like salesforce business analyst 00:23:56 Rajashekar G: basically I'm saying like revenue plus owners link what is my overall growth so here what's happening is this is an interactive graph for me so here this is a pie chart these all are clickable if I can highlight it we can we got information like that's a graph are interactable Whether they made a legend clickable or a bar clickable or a cluster clickable changing the access and why we we for our story needs we have to give all that. So they can able to implement that. So all these three are applicable and this one is completely generated through this user prompt. So user asked all these such a big prompt from this only these are the steps happen fetching dashboard data fetching task counts fetching event counts what are the required uh tables it need to fetch it will fetch and these are all the completed tasks for agent once all this is done this is creating an interactive command center this is entire thing generated in one click like once user enters this prompt and go all these will happen in few seconds and these will be populated and these are all clickable. 00:25:08 Rajashekar G: So these three are link linkled that is the prompt I have given in the as a user prompt here what user definitely want to see where are the clicks he want to add everything and these is all the discussion related because typical CXO right the questions will be the same it won't changes until maybe the question changes by quarterly or uh based on their goals so now need not to come and remember all the prompt there right once the prompt is done they can save the prompt model that can be executed multiple times or they can save the view which will reflect based on once you refresh it. It's like any other dashboard is what and this is all the now what actual uh quick task can you have right now what's the so we need to come up with the design system or design patterns the front end some guidelines at least to get started with the immediate requirement is for Salesforce integration this is the one for Salesforce integration so there is a prospect call with NS So there they want to demonstrate Salesforce integrated with enterprise brand. 00:26:22 Rajashekar G: Now all the demos we are currently relying on this UI which may not be appealing to this is Pyog Python library that is developed by I can override the yes every component will be sitting in there. We can make all the CSS changes everything. Okay. Okay. Now, uh how can we do that? And one more point uh everybody is stressing in every call is we are not saying it as an assistant at least we are saying it as intelligence like Salesforce intelligence whatever enterprise brain int by looking into that structure itself the people feel like it's an assistant kind Yeah. Can I? Yeah. But it is like experial design. That's okay. For example, first idea came for history. So hiding that only if you are giving some option stay whenever you're clicking on it it it took the entire space and it will showcase like a documents kind of thing whatever the pages previous history and uh here search will come whatever you want to search we will search and this will be like a normal uh pages animation kind of thing scroll scroll design exactly the same for examp I don't know whether it will work or not instead of showcasing this one by one by question that you will get only context on that particular previous 00:29:13 Rajashekar G: question whatever graph I'll get more space I'll show this complete graph and everything will be hidden inside but your question and answer How we will do that progressive form just questions they will show all questions will be there but what question you are in present state that only display so much so and combining with the hidden do history kind of thing like how have complete history on the left side hidden up If I want to go back for previous questions or previous versions or then I will expand and I will go back to that previous versions or top instead of scrolling because for me this is a useful for I started with one context of conversation so that conversation multiple scrolls in it. If you're telling an highlighters kind of a thing, here is where you change your context it is not 100% changing but at least field point of view it'll look different and I felt this is little safer also because not major uh change in terms of habituations and all they used to this and all slightly changing few things It is giving a different feel is what I felt and easy to figure it out also easy to learn is also very what I from a new perspective because right now the charging and all your that's leading to an empty canvas it's like and you can start every every new chart every new chart so now with the context we know that which industry which 00:31:14 Rajashekar G: person is going to use and do we need to have a different kind of a landing experience to to start to begin with the one assume that the onboarding for the first time and he's a returning user both the scenarios onboarding for the first time we know the domain we know the industry so can we how well we can represent that the possible questions you might be interested in kind of a thing to begin it's like already the prompt is predefined there so it looks like a question but the inside prompt is defined how to structure that output and all Right. So then automatically gives the answer in a better way. I'm just starting from landing page just breaking the scenarios. So land landing page also how I am thinking that is instead of directly asking that text uh maybe that can be possible if you if you're showcasing history and search together if I want to directly go to my previous chats first onboarding experience split into two different things you onboarding for the first time and you are a returning user both scenarios are different the first time there is no history there is nothing has been tagged to you on boarding. 00:32:32 Rajashekar G: So that is like where you because I I have to understand to personalize your dashboard. I need to understand what kind of the questions that you ask and what data you interested in how you got interacting with the system based on that my experience can be done otherwise it looks like any other we have last access it's not slightly different yes tell me what did we do till now everyone know about And we we we set the expectations right that vision and the scope of can somebody go to the board yogi next time if you are in a workshop if you sit along with the leadership stand there do the writing. Okay. So how are we breaking this down? Where do we start? So learning and all no from we want to start with the UX I mean story we don't want to go with the regular way where do you want to start and some framework okay I have a couple of frameworks in my head there are see what are the questions that this is all about NLP we primarily looking at NLP NLP is NLP voice but there are many other ways of doing it where it is highly driven by agent. 00:34:15 Rajashekar G: So I'll speak randomly. I need somebody to frame this up. I I need you to frame it up. Okay? I'll be here, there, everywhere. But I NPI so AI first products. People call AI first products. There are many ways of doing it. NLP is nothing but userdriven AI first product where user is driving and saying I want this, I want this, I want this. Okay, that is userdriven. The second the second is agentdriven model where it is highly automated and human intervention is here and there. The third one is a hybrid model. I don't know what that hybrid model is at this point. It could be something out of the box. So cyber security that is second model enterprise brain is first model is less. It is all about the system because most of it is automated by correct. Yes. Right. Human intervention is little. You understand that you cyber security understand so AI law I want us to break like that between Yeshuant and Dashik Praak also won't be able to give and we also won't be able to give what are the different types of AI things and autonomous AI some AI right but from a you from a finted perspective where a user needs to ask 00:36:12 Rajashekar G: anatom where automation is the biggest driver. Now in my head I'm breaking it into three models because I usually like three models. I don't know what the third model is right now. Now the interaction models for the first NLP voice it even can be visual video user could feed a picture whatever it is right this is the user inputs first model is userdriven what are the different kinds of inputs is text voice some visuals Okay, each there's a question associated with it, right? For which the system has to give out the answers. This is one. The second model law system needs to constantly provoke the user or inform the user everything is going fine or something is not going fine. Hey, you need to intervene now. Any is the second type of model. Third model will be a combination of something of the both and the TNA will figure it out later. Now the patterns for the first model will be very different from the patterns for the second model. You clear? 00:37:38 Rajashekar G: Yes sir. So now let's focus only on the patterns for the first model because that way we'll also be slightly on the better path. Okay. Now first model law if somebody needs to trust the system it's asking questions somebody needs to trust the system what do you think they will need to trust the system let's start the conversation there end of the day first thing any AI has to do is trust it has to build the credibility and trustwan babu you there you have any inputs You'll share one like which kind of Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: model you using that makes a trust Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: low hanging fruit but okay when I use I think one thing it's doing really well is as it's thinking it's telling me thought process that it's going through process we cannot show the loader we need to show what the AI model is doing that's the second thing progress of the AI model uh because these are all lowhanging fruits for me so high value fruits and then low hanging fruits and the both the sessions that came are low hanging fruits for me Collins UX user experience at a thinking level. 00:39:42 Rajashekar G: If you guys don't change yourselves then nothing is going to happen. Seriously first class next question say they're seeking some information. So information seeking key we already broke it down into this model right description diagn predictions prescription this is your input questions model right input output what are the different kinds of output it can give it can give out voice it can give out visuals visuals can be a picture whatever it can give data visualizations see then let's Break it into multiple things. One is data. It is giving you data. Data is all about some numbers. So output break it into two parts. Okay. Qualitative output. One is quantitative output. Quantitative output data. Some numbers. You're throwing some numbers. Uh quantitative output. Qualitative output and some description. It could be creative description. It could be some description, right? Two things. When it comes to quantitative, it is all about data. Quantity and data. 00:41:30 Rajashekar G: Numbers, right? Data. Now data can be represented in what all forms? Graphs. Tables. Tables. Tables. Documents. Py graph. Documents. Documents. you're dumping something for me. It is data can be represented tables. summary KPIs KPIs then documents GP can data be represented in any other form what are the different worldwide recognized forms or what are the new any new suggestions that are coming up can't be invoke another agent like I just asked for something agent technical problem technicality She just found some process violation while providing data hybrid highly autonomous where the system says interventions third combination of I don't know what that So if we can hyperfocus on NLP righty percent it could be 8020 but if you're considering it as a pure NLP processor what is the ways data can be shown quantitative qualitative quantitative quantitative it's data graphs tables KPIs documents let's stick to quantitative first and I'm just I don't know if there is anything else or not also nothing broadly it's all the kinds of graphs being broken down to different chunks they aren't there So questionart flowchart is qualitative structured process. 00:44:26 Rajashekar G: It's a process. It's quality. There's no data associated with it. It's a picture from AI perspective. It's a PDF document. For example, event logs can also be log whatever it is calling as data just write it down because then qualitative what are the patterns then credibility what is the see this is how I'm looking it okay render circles thank you this is my code qualitative quantitative we are focusing on quantitative so there's one more layer where user friendly Only data scientist will understand regular user any person can understandility. Yeah. Yes. You use claw or something. Different people use different models. Each one of you use different. So ask it what research is being done on data visualizations and apart from graphs, tables and KPIs, is there any new models that the research is at least proposing for the future? You're watching it. If you want tell me summary of research on data visualization we are data spaces and what does that mean? 00:47:40 Rajashekar G: Um 3D scale views room room you can walk in and see like basically Zaxis looking at data points in 3D environment. So 3D data visualization sir interesting data sonification um data sound using sound. So sometimes listeners can detect trends better using anomalies in sound than visuals. Where does that sound when is sound used versus when is visualization used? Okay. But I don't think this experience now you understand neither they'll converge right now. No enterprise bring me the e experiential design building. I'm interested in that one. I think Okay. So when there is continuous inflow of data like for example seismic data or space data related to space or something like that and it's a time series against time you're tracking the same data point again and again it's a real time so then uh two advantages one if you use sound it's more accessible to people say who don't have good vision and also apparently the human brain is more tuned to understanding differences in levels of sound more easily than visual This is like hospital right the pulse and I see whenever it goes beyond a certain threshold deep frequency deep sound based on that the nurses were not right in front of it. 00:50:24 Rajashekar G: So this one applies to real time monitoring and the second one agentriven because it is automated. It's almost like a real time because mostly autonomous is driven by real time most of the time. It's a push push information push where there are real time dashboards where continuous monitoring is happening is where the sonar thing comes into picture. Yes. that so write it down and we'll see if there is any combination that can be taken care like for example they asked me for a descriptive but you also want to show diagnosis so aka we could use things like this where you're going one step ahead in case then we can use a combination of this so that the real information that needs to be called out doesn't get lost. Next, that's it. Can you guys dig a little more deeper in terms of data visualizations and research on data visualizations? Which side the world is heading in data visualizations? What needs to be brought to the table in terms of visualizations? I think this is already something that they said is possible but basically um they're calling it uh AIdriven adaptive view where the page reorders prioritization or shows the information that it believes is the most important for user to see different KPI the KPI that is actually a matter of concern will be placed up top stuff like hybrid 00:53:02 Rajashekar G: model. There's something called scroll telling. Scrolling. Scrolling. Story. Scrolling. Narrative driven design. I'm going you're driving it. Hi Suj. Hey. Hi. How are you? Yeah. Okay. So going back a little before this we were discussing landing page experience and a change that's what my direction but am I telling that that is like in going in one level ahead ahead. So firstly what her direction is very good actually like we'll do this I just I had one idea I just thought I put it out there so that I don't forget. So um at the end of the day we are designing for a person on the other end. Yes. Um so while sees suit person I feel like my problem with dashboards personally is let's say I'm uh I'm a CEO or whatever I get up in the morning when I log in it's the first thing I see and somehow the only thing I'm shown is problems always so that is the start to the day your this needs your attention this needs your attention it is true that it needs my attention but I just feel like for every human being to start their okay with everything that's going wrong is it might not be 00:55:44 Rajashekar G: something they want to do like maybe I as an individual when I come into work for the first 2 hours I want to do strategy building or I want to do more creative things because that is when I'm in a positive space maybe in the afternoon I want to maybe um look at more slightly more deep work towards the evening is probably when I want to meet people. I want to actually sit down with my team and problem solve. So if our platform the landing page can understand that particular person's pattern through the day and then change why should it be a dashboard that's always highlighting some problem really we don't want the dashboard that is the reason what should I show the user on the landing page to provide them context to ask a question so instead of saying these things have to into another is this a good direction for you. I'll come in half an hour. Sure. Sure. But this is good right so now you need to um quantitative now you have certain things how to represent data just focus on how to represent data that you'll go into depth and matter which is one could be KPIs then it could be graphs it could be table is nothing but list view right table has multiple forms to it. 00:57:10 Rajashekar G: Yes. Now take that and if an NLP query comes how will you do it and start off with with especially with AI it could be just one row in G if it's a table it could be 10 million rows yes yes what is your progressive disclosure or how will you discuss there could be many ways you don't have to create the regular type of a list view because the the and 2D interaction do you have to really because what happened is whatever we are doing in a normal 2D design we brought it and proposed it and you're saying go click go click go click I don't think that is the right way of doing it then what is a better way of doing so just focus on quantitative and these and come up with your representation representation what will work what won't work or am I you first question whenever you get a pattern ask yourself am I trying to do exactly what is in 2D 99% of the time the answer is yes for you is it okay to it to be like today maybe yes very strong yes very strong no maybe third one is is there a better way for NLP okay then how will the user consume how will I build the trust pattern is going to go like And I think we have about maximum 10 things that can be done in this space patterns wise for 00:58:49 Rajashekar G: quantitative just finished my thought. Uh so instead of my my whole idea is anybody running a company there's always going to be some problem. They won't walk in thinking today is a chill day. So problem solving has to happen every day. So the chances that it needs to be done immediately immediately are probably maybe low. Maybe they can structure their day the way they want to. So over time if let's say we learn the pattern of the CEO like some people are in a creative space in the morning. Personally when I wake up I want to brainstorm. I'd rather not brainstorm say now towards the end of the day. Maybe some people when they wake up they're groggy they don't want to uh do deep work or creative thinking they'd rather do it at the end of the day. So our um let's say as we were told not to our intelligence if it learns the pattern of that particular user and then the kind of suggestions that we give and the kind of uh work that we push them into if it aligns to what they are like through their day I think then we are being more empathetic to them as a human being. 01:00:07 Rajashekar G: So that is one thing yes that I think we can do differently about the landing page in terms of experience. This is not a visual thing. Second I was thinking this is very odd but um in my head I do not visualize a CEO coming and sitting and typing or coming to one interface and trying to extract. They'll just pick up their phone, call that particular person who they know has the information and say sorry what is going on with this department. So what if our NLP is tagged to a WhatsApp some sort of thing that they can call literally we want eliminate that that's main thing you should not rely on other person to get information for you other person intelligence should be available on call or some sort of if okay not a call maybe an app where the user just feels like you're talking to it And it just happens to have all the information. That doesn't mean that correct also to this right they look at it from a UX perspective. I want you to look it from a tools perspective. 01:01:21 Rajashekar G: tools perspective in a sense like what is the UI okay 2D UI forget it and right now what they're doing is 2D UI if you have to use tool to come up with that data visualizations automate give that additional engagement interactive flavor what are the tools that you would recommend I want that answer from you when I come to this meeting I'm expecting them to propose UX I'm expecting to propose UI and you can look at it from if you want one more team member Look at it from a QA perspective. How would you validate right and I want you and to look at it from technical feasibility technical feasibility allow you I don't know what is involved in to even give you guidance so you and agents if you have to build agent I don't know I just randomly said something because I necessarily you guys won't be able to contribute to anything related to UX right so each one of you from your specialization if you start looking at it maybe we'll get even better ideas you know I definitely want you to look at it from the tools perspective tools first for enterprise brain quantitative that thing look at it from the top But I'm asking enterprise brain question. 01:02:59 Rajashekar G: I don't know what tools can do also. It can't be just about visualization just aesthetics, right? It can be something more. I don't know. 3D modeling sphere instead of pi columns I don't know how you will represent like transitions column I don't know just visualizing that not now if you are asking for me whenever I ask you So until main intent of enterprise is like to eliminate the dependencies and I understand your intent. Your intent is to not to talk to the person. The interface has to give a feel like he's talking to someone. Yes. Correct. To get that insight. Yes. Instead of going to 10 different people, there's one person who knows everything. I want the same thing. And one more thing I wanted to ask you this. So I don't know if this is just something that is possible but difficult to do that which is why it's not being done because I'm authentic. Uh I feel like with them also what will happen is a lot of information will be in meetings right it won't be in um well documented somewhere. 01:04:36 Rajashekar G: It it is coming off of transcripts. It's coming off of meetings. Right now, chat GPT is not capable of actually processing video information, right? I have to take a transcript and then I have to feed it in. Half the time the information is half picked because say there's a screen share going on. They're pointing to something and they're saying what they're saying which my agent doesn't understand. Can we build that? processing videos and audience is in the pipeline format for the next release not in initial version but it is there but I saying is like understanding the on yeah understanding the visual on the video and then connecting it to what is being said will be I think next level phenomenal because most of the time that's the gap we don't need to present the information visually honestly I feel Like again right there different kinds of people I think Rakkesh sa perhaps prefers watching videos really quick ones when I sir I know does I hate videos because you're forcing me to stick to if the video is 5 minute long I have to spend 5 minutes I rather read a fivepage document because I know I read fast I'll scroll through I'll skim through and I'm able to spot the information that I actually care about. 01:05:59 Rajashekar G: So maybe we can also when it comes to the output at some point maybe we can ask them how do you want to see this information do you want me to like right now I can't choose I is whatever it throws at me I'm forced to consume versus if the system is able to ask me I I have processed this I can give you information but please tell me how do you want to consume this that is also I think a change we can make the system should ask the user like which way you want to see this. Yeah, you want I can make a video for you. I can make a flowchart. I can give you a do notebook. It suggest once you have the research data it will ask for would you like to create a flash out of it or streamline. So it's more about based on his query the response can be in one format but you can also ask that like at least with notebook it first gives you a lot of text and all of these other things. 01:06:57 Rajashekar G: Do you want me to create a podcast? Do you want me to create a PPT flashcards are an afterthought like I've already done this now if you also want this let me know instead of that before I give a response if I can invite some kind of um over time maybe I can understand the user yeah maybe the problem with this people right they're adding a too many frictions in between before they see answer that will trigger a different opinion every time you're stopping me in my thought process but all these things are we need to consider But where we are putting for every query that you are asking or whenever certain questions where definitely there is a lot more potential to visualize that in a different format we can ask that or like okay then what we can probably do is maybe a little bit of research will be able to tell us which form is comfortable for the maximum number of people let's say maximum number of people if they're consuming auditory better than text we'll just make that default and then there can be one very small city or I want to make this differently where if the user goes and says I'd rather read about this or I'd rather watch something about this then we change the format. 01:08:08 Rajashekar G: Yeah, we can yeah to begin with we will continue with this we'll first focus on different types of quantitative data and we'll try to along with the visualization of that data and then we'll see what is the representation that we can focus on. Yes, these are the three things that we can do and later we'll see what once more we'll discuss on that first we'll focus on because anyway we started the journey we try to capture Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: that Yeshwanth Reddy Yerraguntla: Uh speaking of uh watching videos with AI, it's already possible u at least for us that we can do it. So here I just ask the question what's on my Rajashekar G: can I see what he's sharing Yeshwanth Reddy Yerraguntla: screen. I just asked what's on my screen. So there was a little setup that was involved but at end of that setup it is able to tell me I'm in a meeting room with several people around a conference table. The discussion is about enterprise brain discussion. Right. 01:09:14 Yeshwanth Reddy Yerraguntla: So what's on my screen the chat can see is something that is technically possible. Rajashekar G: Okay, great. Yeshwanth Reddy Yerraguntla: Yeah. It's not a video aspect of it like that. It it just sees what's on your screen. That's all. There's nothing specific about a video or anything else. Rajashekar G: If it is a video it will convert it into frame by frame then it will read it out. Yeshwanth Reddy Yerraguntla: I Rajashekar G: Same thing we did for visual detection of something. Yeshwanth Reddy Yerraguntla: mean Rajashekar G: So the real understanding is a video will be provided the video will be uh cut down into frames to frames frame by frame then it will be loal and in the future I feel like that's a I don't know if you can call it a market gap but that will be a USB if you're able to actually process videos with the context and understand what is actually being said even now it's more it's just that the size of the very I don't think it's for example at least not that I can see Anyway, okay. 01:10:45 Rajashekar G: We'll focus on Rag. How do you want to work? Yeah, same. We'll continue with that uh the quantitative data, different types of quantitative data and how we are going to visualize those those data points and now we want to represent that now because typically the KPI can be represented as a single number and key and value. Is there any better way to represent that in a One thing I was thinking on the experience standpoint which you have also partially conducted the same thing is that uh considering the seesuit members if it is targeted for that they are on a different context for different meetings. So the agent will be able to skim through their calendar and find out and accordingly update the dashboard dynamically based on what they wanted to say. Exactly. Because right now we are one step before that because once the assume the dashboard is changing dynamically based on the context assume that there is a modes we can able to switch the mode the context is changing. Now we are talking about what are the different ways of he prompting to the system. 01:11:51 Rajashekar G: He can give that for example a count as an input to it or he can give the log as an input to it or it can be meeting minutes as an input to it. What are the different types of quantitative data? Right now we are trying to identify what are the different types of quantitative data and then visualization. Then no matter what in whatever mode he will be interacting with the system in any one of this things only can can we also learn and personalize the dashboards like don't go to the dashboards at this part because not dashboard something that you are representing as an output should be suiting to my personality. The system should know what I am, what am I, what I look for, what is more appealing to me. that we'll consider in the step two because otherwise right we we will be thinking about the experience now but the core is like intent is to first understand the core of data because maybe this can be used in this or for example assume that this is hyperpersonal system we build we know that based on the role and your intent the system will changes but we are going in a direction of a dashboardless interface because on the day we discussed about those things and we don't want to come up Start with a dashboard to fill in but once you ask a question that will create a dashboard for you to further drill down to get more insights and more further questions. 01:13:14 Rajashekar G: So first we'll try to focus on at least we'll have something to uh continue our discussion. Can we create a quick Excel sheet and we'll put the different text there. Okay. Um I'll start screen sharing so that everybody can Uh what do you want me to the data types? Whatever kingdom whatever you got different types of data we'll add those things those things and also share the sheet with everyone so that we can also come to check. Is that you? Yeah. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: will be using internally HTML and CSS completely HTML Uh we Yeshwanth Reddy Yerraguntla: HTML CSS or main pylocks. Rajashekar G: might Yeshwanth Reddy Yerraguntla: Uh so by the way we whenever we use this term pylog it's it's a library that um is a UI library that uh uh we named it uh because it is written in Python and supports dialogue. So we I just sandwiched both of them Python dialog into pylog. Rajashekar G: And that is generated by you Yeshwanth Reddy Yerraguntla: Oh no no no like everything is in my control right now. 01:15:56 Rajashekar G: completely. Yeshwanth Reddy Yerraguntla: Mainly it uh communicates via this markdown syntax. I don't know how many of you know what markdown syntax is. But uh you can assume markdown is nothing but a replacement for text syntax except that it has some more uh functionality where it can support bold text, italic. Uh it it has some nice uh CSS for table display. End of the day chat is text. Markdown is somewhat fancy text. And um on top of that, Pylog currently also supports generation of charts uh in the chat. Rajashekar G: That's Yeshwanth Reddy Yerraguntla: So when I say give me the Rajashekar G: enough. Yeshwanth Reddy Yerraguntla: fastest chart. Okay. So currently only markdown text and charts are supported. I am trying to do something u different with it. I like I like just woke up and uh had a call with uh Navi Nana. So, I'll just tell you what I'm trying to do on my screen. Rajashekar G: One minute. 01:17:27 Yeshwanth Reddy Yerraguntla: I'll just Rajashekar G: Switch to this. Yeah. Yeshwanth Reddy Yerraguntla: Yeah. Yeah. Yeah. I'm sharing. Is it visible? identity. Rajashekar G: Let me take Yeshwanth Reddy Yerraguntla: I think I just stopped this thing. Rajashekar G: question. Yeshwanth Reddy Yerraguntla: One second. Yeah, can see my screen. So, so currently the interface looks something like this where you have uh the history, you have chat one, Rajashekar G: layout. Yeshwanth Reddy Yerraguntla: chat two, whatever. Rajashekar G: This one Yeshwanth Reddy Yerraguntla: Huh? Rajashekar G: I'm just Yeshwanth Reddy Yerraguntla: Leo Leo uh I want to tell you something that I'm trying to do which uh I'm which I just need your which I think is possible and I want you to just empower with that ability as well that AI can already kind of do this. So think in that direction kind of a thing. So currently whatever chat interface um whatever you speak something it will say back something this is all fine I'm saying currently it can also generate charts right so this is X this is Y it can do this on the 01:19:05 Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: fly um only This Rajashekar G: Make that 2D graph to 3D Yeshwanth Reddy Yerraguntla: picture Rajashekar G: graph. Are you using Python code to generate it the graphs? Yeshwanth Reddy Yerraguntla: either currently this is Rajashekar G: any graphs, any data. Yeshwanth Reddy Yerraguntla: Python. Rajashekar G: That's what I I want. That's what I Yeshwanth Reddy Yerraguntla: So when I ask something it is not generating the chart by itself. Rajashekar G: asked. Yeshwanth Reddy Yerraguntla: it is sending to uh Python what it thinks it has to generate. Python will generate that chart and put it here because this is far more uh reliable. LLM by itself it will hallucinate a lot when it tries to generate charts and on top of that this can handle uh you know thousands of uh rows also right it this chart is going to be very accurate accurate in the sense it's perfect as long as LLM can tell me what is the x-axis what is the y-axis what is the chart type whatever specifications it gives as long as these are correct this will be correct Sorry. 01:20:29 Rajashekar G: Thank Yeshwanth Reddy Yerraguntla: Uh this is what I'm planning to do at least. Rajashekar G: you. Yeshwanth Reddy Yerraguntla: Uh here what is going to happen is the user will ask it will continue the chat but I have a blank canvas here where when it generates a chart and Rajashekar G: Open. Look Yeshwanth Reddy Yerraguntla: layout let this be something else all together I Rajashekar G: at Yeshwanth Reddy Yerraguntla: am saying I want the LLM to I'm going to give the LLM the ability to say okay this I put it here. Okay. So the tomorrow when I wake up maybe I already have three different uh these are not charts by the way. This can be anything. This can be text. This can be a document and this can be something else. One of them could be a chart. But LLM I'm saying will have the ability to if I ask what is this element? It should be able to read and say okay this is what you're doing. 01:21:33 Yeshwanth Reddy Yerraguntla: And I can say change that element and it can change it. Add a new element and it can add this kind of a thing is what currently I'm thinking of uh enabling in pyog just so that you guys can think of more powerful interactions. This is not a dashboard as such. This is just building the web page on the fly. Elante thought processes thought process I want I want to consume from you guys also beyond going just beyond chat interaction Rajashekar G: I know when I'm doubt something like 3D elements comes in. Can we able to generate that using Yeshwanth Reddy Yerraguntla: if this is once this is done yeah what will happen is Rajashekar G: pyog? Yeshwanth Reddy Yerraguntla: um if there is a 3D element element, you will come and put it here. One more 3D element, it will come and sit here. Or this can be one 3D, this can be one 3D. Each of them is an independent HTML um thing where as long as you enable it with enough JavaScript and CSS, it should act as a standalone uh we can call them widgets. 01:22:56 Yeshwanth Reddy Yerraguntla: So there's no restriction saying I can do only this much in each Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: widget. Anything should be possible. Yeah. So I just wanted to share that we are currently trying this kind of a activity just to break out of the chat. Yeah, Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: I mean I don't know if I it was my place to show this but uh I just wanted to give my current thought Rajashekar G: Instead of just Yeshwanth Reddy Yerraguntla: process. Rajashekar G: chat Yeshwanth Reddy Yerraguntla: So widgets may operations for the UIP. Rajashekar G: direction. What exactly are you trying to do? You're trying to say that these are all the types of data that are there and then what? Now what is the way to represent the data? Right now in how we are going to represent that table not related to this only table of events data types quality data types there are 10 different types of things in the areas and the uh continuous sequences in the we have a almost 10 or 12 data types are there which is falling under the quantitative now each type there is a way to represent right now in the how we are representing right now considering the AI how we can change that representation okay and the third step first 01:25:56 Rajashekar G: step already we did an event itself we identified what are the different quantitative data type now we are seeing that what is the way how we are representing right now. The next is like how we going to change the representation. These are the things. So uh vibration patterns that can also be one of the Yeah. Yeah. What is that? colored agent may have already analyzed the data and segregated them into product classifications. each of the find out that there's a brief summary with each person on top of it I can ask question even for the time lapse also if I ask a Yeah. I don't have to write much already. So it has already segregated the information that I look for and I have the information This candle stick motion. I can. Okay. Yes. Let me just keep this back. question. Good. The fruit. Qualitative quantitative interesting All right. 01:47:45 Rajashekar G: You can't check, right? Where are you going? I don't need you to sit there. She asked you for a scope document. Who did ask question? What the hell are you talking? India is saying she didn't ask. You are seeing you. What is going on here? See they asked after we sent them as no during the conversation with Is that what it is? I need you to go confirm with and tell me what happened. You guys are figure that out. Okay, you can go. Are you waiting only for me? You said to connect for the invoices, it stories you guys. Come sit. I'll get my laptop also. You can show them. There you go. We'll connect after this. Meanwhile, I can pull something out. He's in hospital. He saved me in hospital. Yeah, he's went to check up. Whenever somebody says I'm in a hospital, just ask what happened. 01:52:01 Rajashekar G: I know the reason. Normal checkup, right? Yes. Also, he went for that insurance claim or something. Huh. So, you have to just say he went for a checkup or something. You saying he's in a hospital. You sound like you know something big happened which already happened. Go. Where are you? These So do qualitative data visualizations tab quantitative qualitative make quantitative or qualitative both I made separate this is quantitative this is quantitative I can't I And I don't want to continuously keep telling all of you I can't read. I really can't read. Make it further. That's enough. 150. 150. Yeah. What we have done what are the different data types and how traditionally those are represented and with the a first approach what how can this is presentation uh the quality more than quantitative. I like the qualitative data validations because we feel like that we are doing it but we can focus more on the a first era. Well, first let's focus on quantitative because the enterprise brain flow uh insights is all about quantitative. 01:54:03 Rajashekar G: You want to drive it from go ahead. Yeah. Can you go there? For example, the nominal letter typically we represented in a table format or a list views supporting tickets labeled by assume that we the scenario is like we're looking at different support tickets with whether the labeling or a status we take talking about either building or login or delivery related that is a traditional way we represented with a simple tables or a bar by categories sharing screen you not sharing just Okay. Can you increase the size here also? 150. He increased the nominal data. For example, I'm taking so to get more context taking a scenario for now here scenario is like we are looking into a support ticket listing page. For example, when I ask for what are the open support tickets that user made then typically in traditional way we'll give a simple table or a bar chart by categorization. Now what they are saying is like in a first we can tag instead of showing a simple table we can tag it to a semantic tables. 01:55:19 Rajashekar G: What actually the semantic tables uh is about uh it's more about qualifying the instead of calling it as a tax we are qualifying it because of uh low engagement tagging to the semantics and I mean as far as I'm understanding it has to be what we did in core dashboard with the graphs that is semantics in my so for example that is categorizing like a billing refund we have to map it to the it's a You want to share? Sorry guys. This is an additional way of representing the raw table. If you assume that this 101 is a building category and all that in case of semantic table we are going to categorize that it's a billing is related to the payment issue and the reasoning can able to and uh for login it's an access issue they able to categorize that That is not a UX pattern. UX for example you remember science insight. It was a combination of a mini graph quantitative data categorization. Yes. So that is what comes in the table. 01:57:24 Rajashekar G: data models data inferences and insight I don't need is what I'm rule, for lack of a better word, there's a rule or there's an AI. So this one even a rule engine can give me know data one is AI we still user experience perspective user user experience perspective they are the same in fact we did even complex list views complex Because of the B2B requirement, we were forced to do that. To the degree that we did interactive graphs, an entire new gamut of information opens up. Nobody would even assume that we designed. Truth be told, manu have done patterns that I we did really fantastic patterns uh in terms of just the UX and you're with me or you're still confused and the exercise that we did what is the traditional way what is the qualitative and what are different quantitative datas and once you do that So it is one more column you added unless you come up with a different representation. Okay. If you're saying there is a semantic layer that you want, you're adding on top of the regular information. 01:59:30 Rajashekar G: I need a different representation of the semantic example semantic table. We didn't even crack it. Oh, okay. You're just saying that because justice is like okay then I assumed I assumed the design is correct and we are telling other okay then I sorry my bad I got confused then I thought that you're proposing this I as able to differentiate both how to do it we haven't for example But majorly do you see any difference between um ways of representation only different but I feel like it's interesting exercise Because so many are there which we will do but we don't know what the actual data type is then can you take this to the next level R do you need me not really next step is to how to represent it now whatever I have told you if I am using an NLP based system. What is the better way to represent that each and every workshop problem? So who will own this? You will own this. Can I trust you again? 02:01:49 Rajashekar G: No, you'll be okay. Update. We have a move. We have to make everyone work that. By when? Give me a deadline. By when will you come up with a design system for this? It has to be within the next two days actually. Now work backwards and tell me within the next two days by not all. We now need to work yog I want you to own it and you need to call people. You need to set up. You have to order food and make people sit. I don't know what Vinutna will entice V now. I don't know what will entice me. If you have to call me, I have another 10 priority things that Navin is waiting near. So I need to pick you over Nin in terms of priority. How will you how will I pick you over Ninput? Yeah. If I have to start off, I mean I already I'm talking already. It's interesting to me as we so focus on this. 02:03:19 Rajashekar G: I will you continue this workshop then. Are you guys okay with one more hour? We'll identify the real the top top 10 patterns which are commonly used in this list. Qualitative will try to prioritize. You want to go with qualitative. Most of our outputs are like I am connecting to this other one is like a simple KPI and all that. not saying that then it not connecting. I need him to connect UI for there's not much UX and already there hand it out to V and asking show me how you're going to represent that from a UI perspective take it over you work separately with don't don't be part of the discussion so you know the patterns patterns I will come here anyway knows the patterns maximum the patterns then tell me how you are planning to implement them and what is the tool that you are going to use to do that qualitative I'm not seeing mostly graphs so single qualitative where it's like a more interactive things are coming the interacting I ting Florida. 02:05:04 Rajashekar G: Florida. So it is giving a speed control also for speed control side you have the data and individually because the entire gamut Okay system you won't get any excitement there just a visual st but see a board and other pl the cluster and you want to see the cluster because it's a 2D plat visualize where you have a three-dimensional access to look into then you have to bring the 3D so they have a good line so all type of patterns covered in terms of graphs and it It's interactive also and some of the graphs are having like a deep level drill down also and the seven step or five steps each uh the transitions also happening. Okay, we'll do one thing when I uh we'll create five different scenarios. What are the typical questions? First we'll identify the pattern which we want to work on here in the qualitative data type and the quantitative data type. Assume that we are picking five from here and five from here. Now once we identify the pattern we'll try to generate a question for that in order what kind of question will lead to that kind of an answer and we'll work on the representation. 02:06:55 Rajashekar G: So first we'll pick the qualitative and quantitative and assume that everything is interactive even qualitative data can be interacted and quantitative data can be interacted but there we need to see for me just a data representation and all that but ultimately it's leading to either in case of quality always a graph a single visualization where in case of here it's a table or a cluster data or a dash it's giving to because it's not able to answer yes or no kind of a thing is always leading to an answering multiple questions at once. I have found out like recently about drilling down like if I clicking on something other chart is dependent on this other table. It is changing but what I heard is on there is another prompt or suggestions coming from How well we can represent student. Sorry. Go to the column A. Very simple. because this is what we wanted to present. You can start here and example just delivery late app is slow they are putting in clusters but it won't be a strong scenario feedback is coming for example my visual detector different types of things and you forming a cluster on that on top of that how you are creating an evidence table for that one is cluster representation no now you have a weight to drill down also one cluster which will go there so representation we are working only on that representation right now uh support But at the end of the 02:11:06 Rajashekar G: day, there is no data. Semantics particular three word groups carry four word groups. based on Ukraine. Yeshwanth Reddy Yerraguntla: Hello. Rajashekar G: Hello. I'm muting Rakkesh Yenugudhati: Hello. Uh, can you hear me? Yeshwanth Reddy Yerraguntla: Yeah, I can hear you. Hello. Hello. Rajashekar G: I can. Yeshwanth Reddy Yerraguntla: Uh Rajar. Rajashekar G: Hello. Yeshwanth Reddy Yerraguntla: Yeah, I can see the screen. Rajashekar G: iling just calm. email lead. I think was you also friend Yeshwanth Reddy Yerraguntla: Yeah. Yeah, I'm here. Rajashekar G: on the clusters are a dynamic and it's a question on top of the question. So there are it takes that context and it builds everything. So a models on a chat if you're going in depth and even the tokens that are consumed. Yes, you're here right I'm speaking on your behalf Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: explain this is not my knowledge this is knowledge proprietary information okay chat local if you're digging in for every question the tokens that will be consumed within that 02:30:08 Yeshwanth Reddy Yerraguntla: H. Rajashekar G: dig of it is much more expensive than a new chat okay drilling Yeshwanth Reddy Yerraguntla: Yes, Rajashekar G: down drilling down is more expensive That's what I understand. Yeshwanth Reddy Yerraguntla: correct. Rajashekar G: Can it be done? What is this? It's just a framework Yeshwanth Reddy Yerraguntla: Just a framework. It's just a visualization chat framework. Rajashekar G: combination. Pilox, please. Yeshwanth Reddy Yerraguntla: I'm curious what do you mean by can we do this like this graph we can show tamund you are showing something else right where you are showing A journey of clusters that Rajashekar G: open show all of this. Okay. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: These are some of the examples because without this Yeshwanth Reddy Yerraguntla: H. Rajashekar G: You already showed me that some of the interactions are possible but it's like a a complete it looks like more more modern chart library. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: See either we use a third party or we build these models uh as Dwami library right and on top of it what else can Diwami do on a proprietary trigger we we have that and we install and use that okay where we are trying to save money and not using plot which will charge you for every damn thing right this is one approach I'm just speaking out luck but whatever you guys are suggesting existing is this pylock free framework or it is charged framework it 02:32:25 Yeshwanth Reddy Yerraguntla: Free on. Rajashekar G: is free framework free now can we do this in that is my first question if you can't use pilogs for this can you find another framework I don't know where he's showing all of these these are the things that I'm talking about we need Yeshwanth Reddy Yerraguntla: Hm. Rajashekar G: right then if not some somehow we need to figure this out Okay, this is one part we need to be able to do. On top of it, Rakkesh and team will come up with more maybe that's the second point. Third point as of now NLP if I have to do this everything is like either a new chart or so we need to figure out what is the related how do we build that navigation how do we build the structure the hub and spoke model bigger structure so as of now for example in our other things hub and spoke or models models What are other things that are involved? Experial designer. I just Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: message. Yeshwanth Reddy Yerraguntla: You put him in there. 02:34:41 Rajashekar G: Huh? Yeshwanth Reddy Yerraguntla: You put him in there. Rajashekar G: Please move. Hello. I don't know. for mention for example Yeshwanth Reddy Yerraguntla: Huh? Rajashekar G: Okay. It is not important. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: Do you have such an Yeshwanth Reddy Yerraguntla: Yeah, Rajashekar G: ideal Yeshwanth Reddy Yerraguntla: but I was hearing multiple things. Yeah. Rajashekar G: which is called when it needs to render and type. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: So the same kind of functions if you do have something like we Yeshwanth Reddy Yerraguntla: If this will give JavaScript framework Rajashekar G: JavaScript Yeshwanth Reddy Yerraguntla: No, Rajashekar G: framework. Yeshwanth Reddy Yerraguntla: I don't see why it's not possible. Rajashekar G: Okay. 3GS code. Yeshwanth Reddy Yerraguntla: No, Rajashekar G: Same thing we can also do Yeshwanth Reddy Yerraguntla: no, no, Rajashekar G: here. Yeshwanth Reddy Yerraguntla: no. Those are already flourish. Studio Rajashekar G: Yeah, but now you not sure what you This is what I want to achieve. Yeshwanth Reddy Yerraguntla: Yeah, at this point uh based on this activity, we'll have to decide whether uh pyog is the right way to go forward or we have to build our own thing in react based on the uh demos that Raja just showed on screen. 02:39:21 Rajashekar G: English Yeshwanth Reddy Yerraguntla: I was saying technically whe technical decision whether you have to go forward with pylog or uh some other framework uh and Rajashekar G: in the killer to begin Yeshwanth Reddy Yerraguntla: because everything will be in Python it is faster to write aent code and debug what is going on Rajashekar G: But it won't be rich in user Yeshwanth Reddy Yerraguntla: react it is currently Rajashekar G: experience. Yeshwanth Reddy Yerraguntla: doing bare minimum stuff that it can show charts, it can give the user uh input response Rajashekar G: But this is not our strength from a UI perspective. Yeshwanth Reddy Yerraguntla: output anta. If we are going if you're agreeing that we have to do far more complex uh visualizations, we have to take a Rajashekar G: Yeah. Yeshwanth Reddy Yerraguntla: call. Rajashekar G: Yeah. I want far more complex visualizations. That is one thing I've been trying to tell all of you since day one. And this is what I've been talking. This is what I told. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: This is what I told. 02:40:40 Rajashekar G: Okay. to you especially when you were coming up with the UI designs. How come you these didn't come up in those things? Why is that the designs are done like that? more designs actually based on the that's directly implemented I mean generated area designer yes a list view with page you guys did it without even consulting I don't have the context when are you speaking who is the context And you let it happen. I need Rakkesh signature. It has to be Rakkesh signature or Pratima signature. I'm not even giving because you want your job to break his design. Okay. I execution. I want you to start showing that this can be done better. That can be done better. UX will drive you. He needs to identify all the patterns and give us a pattern. Then you need to figure out which pattern will apply where and screening here will be more like color usage and uh circle shapes that these guys can write configuration files in AI that will automate Yes, ma'am. 02:43:36 Rajashekar G: little size of the circle is configurable. A circle circle can change into a diamond is configurable. Shape can shape is configurable. Size is configurable. Color is configurable. You won't need a UI designer anymore for that. We are depending upon the UI designers and configuration whenever we ask I don't know how you guys are going to do it for the same with me right I'm sharing the No. Yeah, I see that. Yeshwanth Reddy Yerraguntla: What is up? Rajashekar G: 10 minutes. So I don't Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: know if this is product engineering I don't know whether it is I don't know and I'm talking to all of Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: you and all of us have understood this is the next we have to do Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: for example flow what are configurable the shape the size of the shape and color and the color yeah and then when you hover the configurations Right. Right now you're only showing information. It could be like multiple states associated with it. 02:45:48 Rajashekar G: Now this one is still involved to do this separately and do this separately. I don't want a UI designer to be involved anymore because those things can be automated in my head and that has to be driven from a configuration. How you are going to configure? I'm not talking about the approach. I'm saying it has to be it has to go to the next level where it just takes a couple of inputs from somebody as good as a UX designer or it could be you know a developer and then take it forward from there. Okay. All of you have one big problem. When you are listening to a requirement, why are you trying to fit it into a solution that you are aware of? All the confusions that you guys talk about but still hazy is only because you're trying to fit my requirement into something into what you know. If you already know why would I even give you a new requirement. Got today's maturity though you won't be able to solve my problem. 02:47:17 Rajashekar G: That is why I'm here. Right. Got it. Right. So don't try to fit my requirement into what you already know. Do it later. So the step I would do it is first listen to the requirement completely. While you're listening don't use your brain to process it. Sure. Process it in the sense of solution. Listen to the problem. Okay. Now you need to once you listen to the problem I need all of you in this room to understand this part of it. Listen to the problem. Once you listen to the problem, ask yourself, is this guy really talk about it as a problem or talk as if this is a solution? I actually talk to you as solution. If you realize, okay, then you should not accept the solution, then what's the point? Why are you here? If I am giving you a solution, right? They will not value you. Got it? Got it? 02:48:11 Rajashekar G: Because if I am giving you as a solution, if you're just executing, you're just a doer for me, not a thinker for me. So I bloody will not give you that much value. You understand? Second point. Yes. Now whenever I talk to you in terms of solution, convert it into a problem. What is she actually talking? What am I talking? I'm talking about building an automatic design system. Right? That is the actual problem. That problem is it just that a big problem that has to be solved or can I club two three problems and make it an innovative problem. That should be your fourth step. Take that innovative problem whatever that atomic unit of entity or problem is and then dissect it and say how can I automate it or how can I apply whatever this is how you need to approach. Are you all clear now in the fifth step is where is do I have an existing solution in my head that will solve this? 02:49:11 Rajashekar G: Should it be a combination of multiple solutions that should solve this or should I come scrap everything and come up with some other approach? Yeah. Our brain will only give us solutions which we have seen in the past. I want all of us to understand this. Okay. None of you thought about those patterns in the good because it will always go into the past and check what is in it and then according so unless you push yourself push yourself. Why is depth of these workshops so important? Unless you go deeper and deeper and deeper and deeper and deeper, you won't come up with these ideas. So whoever is coming up with these ideas, they give that time to it. They let it just but that gestation has to go depth not parallel. explain. I'm hoping this is the last time to at least explain to anyone of you in this room. Okay. What next? uh uh see I don't know if this has to be 02:50:39 Venkatesh Tammareddy: Okay. Rajashekar G: into an AI problem or it can be strenuous is also there and Rakkesh is also there. Okay. Venkatesh Tammareddy: Okay. Rajashekar G: What I need is can you can you close how it is uh grouping all of Venkatesh Tammareddy: Yeah. Rajashekar G: them? Venkatesh Tammareddy: Yes. Rajashekar G: It it can it be a standalone front end component that the AI Venkatesh Tammareddy: Yeah. Rajashekar G: team can consume and even you as 2D designs can consume or it has to be will it be tightly coupled with AI between Neshwant and you I need answers Venkatesh Tammareddy: From the visualization's perspective, it is completely pure front end Rajashekar G: can youulation From perspective, Venkatesh Tammareddy: account. Rajashekar G: it is completely fantastic. Okay. Yeshwanth Reddy Yerraguntla: But the important thing here is um what to visualize needs to be decided by AI. Rajashekar G: Classify data and categorization. Which pattern to use has to be decided by Yeshwanth Reddy Yerraguntla: H correct. Rajashekar G: Yeshuan. Yeshwanth Reddy Yerraguntla: Oh Rajashekar G: Then where will design come into Yeshwanth Reddy Yerraguntla: no. Rajashekar G: picture designer will have a lot of grip on the uh data where is coming from is AI is an evolution will keep I mean at least correct me if I'm speaking something wrong. 02:52:52 Rajashekar G: So the model keeps evolving itself is what I understood depending upon statistical model even LLM the data that it is receiving. though a context in which this one may keep changing slightly different what I understood is now we're talking about interactive graphs right so I think said that how we are showing the animation everything is purely front end has nothing to do with what is saying is what question such an is what I you are absolutely right. There's a small tweak here in our existing applications assume that the call from the front end to the back end is and back end is giving you some information. How to represent that information independent of how much information is there? We already picked one way of doing and this show it as a column graph and as a designer we would have given it already. Right? Now independent irrespective of what it is how much data is there if it is a little bit if it is nothing you just show no data if it is one column you show one column if it is two columns you show two columns if it is like million records then only your x-axis and y-axis will change but the context is around that graph and that graph would have been defined by the uh designer saying that this is a better data representative graph for this particular 02:54:29 Rajashekar G: thing. Yes. Okay, I'm going one level above as to what you're shown this because the data keeps evolving, the choice of graphs also will be dynamic and will be changing. Yes. Okay. The when you say data keeps evolving you mean within the context the volume of data the volume of data about virtue AI is all about as of today with considering this data this is what I'm predicting or this is what I am doing now the volume of this data increases then they keep the model will keep evolving and the number of edge cases and all of that also will keep increasing You're saying the visual visualization that works today with the current set of data might not work as the data changes it may be heavily datadriven rather than userdriven that is where I keep saying it's AI first AI first and AI first law see there are certain things that have to be consistent right typically user experience and then what what needs to be consist consistent and what can be dynamic. This is what is one of the user experiences. 02:55:48 Rajashekar G: Right? Even when we say innovation, innovation from one product to another product is innovation. But innovation, look, what are the patterns that you're defining end of the day are consistent. So when a user comes to a particular place, he knows that are consistent. So there are no surprises for him in terms of patterns. There surprise to him in terms of consumption of the data. You understood the difference in traditional ways we need to be we need to understand this is where the designers and design firstly I constantly and continuously keep saying that what the fundamental philosophy of a 2D design does not apply to this I constantly keep saying that what is consistent and what is dynamic in 2D varies a lot when it comes to air first. So what should be consistent? What should be dynamic? You need to give that freedom. You pick a table when you have 10 rows in max 10 rows but beyond 10 rows. Now as an example that AI threw 10 rows. 02:56:59 Rajashekar G: So you picked a table but tomorrow through 100 rows because of the evolution of the data however for now if you still pick table there it'll be lost. This this is a perspective I'm coming from. I'm not saying that fundamentally right. All this I understand but I don't have I don't have guard rails. I don't have frameworks that in my head at least based on which we can give a direction to our intelligence. That is what this workshop is what I'm expecting and we are defining guard rails. You're defining a direction cases this is how we need to learn cases. So that is what end of the day is frameworks right? Yeah. For example, understand the context the AI has taken a first step because considering the volume of the data and it tells that this is the right way to represent this and show the visualization to the user. Maybe the user is not comfortable with the visualization. So because you feel like the class clustering is very useful for me but I felt like the clustering is not I'm able to consume because it's also depends upon user to user based on how what he can consume because I'm a very uh logical guy I want to see in the table format for my analysis and all that so I can understand it what I am doing even though I'm sitting at the same level in the different guys but I prefer more certain visualizations in my questionner as a response some other guys prefer that something else so system can also 02:58:38 Rajashekar G: mature the pattern of that the usage that I'm in the sense right if there are 10,000 records the whole point of AI is it does it it it digests everything and it spits out something very simple for me now if somebody wants to go and drill down they'll only want to drill down per what and a is what we need to understand and crack that and give a better way rather or you need to give him a downloadable report maybe okay and okay let me rephrase for our understanding if you are saying my normal 2D pattern would have been pie charts graph chart which we have done a lot of times semant lot of them you want a different kind of a representation Nigama we gave the representations where a user is driving it. Yes. AI is driving and you can't you can't say you have to tell user you're dumb. AI is more intelligent. Twinkle old. That's what AI first is all about. Right? You're dumb. I know what you need. 02:59:54 Rajashekar G: Here is the thing. You tell me what your problem is. That is a solution already. And if it gave me as a 100% not that one I want like numbers I want numbers user is driving the solution. If ever to drive the solution, you need to understand the problem of the user which is now pi is always like percentage for me. No, I'm I I want a breakdown of something and I I hope you are understanding where I'm getting. I'm probably may or may not be able to express it. AI should drive it. So you need to understand the intent of the user intent intent of intent and air should be taking those calls is how I looked at it front. I am saying AI first. So AI person AI should be driving it. User is secondary in the sense like you have to tell the user I'm much smarter than you. Let me decide. You just tell me the problem. So we have to think differently. 03:01:07 Rajashekar G: Yeah. Yeah because few other articles that I'm reading because right now V is taking the control of completely the people are not trusting the systems we come to that next the third rule okay I'm not saying to two things I'll tell you one is trust experience. We are doing this. So what I mean by this is you need to give an experience. How will you give this experience users to feel they have they have control but you have to build a trust that three different things. Yeah. Okay. Control doesn't mean you are asking user for a solution. Control means user should feel like I should be able to ask any question and AI should give the answer. Okay. Control was that way. better solution. So should give a better solution. But trust is different, control is different, experience is different. Just because you give a great experience, I will not build trust on you. Giving you 100% trust that AI is working really well doesn't mean that they have great experience. 03:03:07 Rajashekar G: Control is still making it interactive. Immersive and interactive. Immersive is highly experienced driven. Interactive is highly controldriven. That's how I looked at it. I could be wrong and it will all depend upon our workshop and the depth of the workshop. Okay. So you know where we are heading. You know what your next project is or what's my next project enterprise brain. Your next task. What's your task? Are we capable enough to bring this into our entertainment UI? I'm not asking you are you capable figure out how to bring it is what I'm asking. Yes. Okay. You are capable. That I'm not text you between you. I included you because you are the printed expert right? So one independently this can be built. I got I know that but ideally I would have included proposal that they working on. I didn't want to disturb him. If we have to go back to him at a little later or tomorrow first thing we'll go back to him and say that this is the direction we are heading in. But let me put it this way. 03:04:51 Rajashekar G: I don't know the next steps between you and Venkatesh Tammareddy: Sorry class. Rajashekar G: I and then next steps honestly I don't know where product engineering ends and where AI comes into picture or where AI ends and product engineering comes into picture. So I between you and Ashwant can you guys tell me tell me Gadu what is the next step? Oh, Venkatesh Tammareddy: because I was in middle of it. I understood only a little bit. So now I need to have a clarity on what exactly we're trying to solve. Then I can work with Ashwand and come up with the whatever is required Rajashekar G: me and three of us have the whole context. Venkatesh Tammareddy: and Rajashekar G: Sure. Venkatesh Tammareddy: exact. Rajashekar G: What can you want? Do you want to get into another meeting and discuss and open it? You don't need money. She can leave. Venkatesh Tammareddy: Uh Rajashekar G: You need Venkatesh Tammareddy: so Morning Rajashekar G: money. Venkatesh Tammareddy: connector. Rajashekar G: Okay. Venkatesh Tammareddy: Uh maybe it will be an hour's work actually. 03:06:33 Rajashekar G: I don't know what this is. It may take Yeshwanth Reddy Yerraguntla: I think we should spend like 10 15 minutes to understand uh what are the expectations and then we can meet tomorrow in more Rajashekar G: Yeah, Yeshwanth Reddy Yerraguntla: depth. Rajashekar G: I would like that if we have 10 minutes, I want us to understand what are the expectations and what each one of us is doing so that we have clarity from a UX perspective. You didn't have clarity like you're still continuing with your works. You can go look at these kind of patterns and understand. You now need to map the patterns. User first predictions. Enterprise. Pogat Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: itself agent library we have built you built. Yeshwanth Reddy Yerraguntla: Heat. Rajashekar G: framework with UI and back all is a third Yeshwanth Reddy Yerraguntla: Pilog we wrote it from scratch by Rajashekar G: party. Yeshwanth Reddy Yerraguntla: ourselves. Oh no. Once I understand uh how to make JavaScript uh Rajashekar G: Got it. Python based JavaScript back 03:08:32 Yeshwanth Reddy Yerraguntla: talk I think it is Rajashekar G: code. Yeshwanth Reddy Yerraguntla: possible to like once I need to sit with Watana and understand the technical aspects of it but I am not uh saying this is impossible it's just that how you Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: structure the architecture will uh Rajashekar G: Third Yeshwanth Reddy Yerraguntla: Uh Rajashekar G: party normal human on earthmot Yeshwanth Reddy Yerraguntla: yeah. Um we wrote it mainly for Rajashekar G: foreign. Yeshwanth Reddy Yerraguntla: Yeah. No. Um, Rajashekar G: Yeah. Yeshwanth Reddy Yerraguntla: so currently the capability is like I asked some question like uh Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: um show me this visualization blah blah blah. It will run a query Salesforce. It will run a query. It will decide what is the right visualization for that query and it will show Rajashekar G: Okay. So, Yeshwanth Reddy Yerraguntla: it. Rajashekar G: so there are two aspects. Can we can we come up with steps within the next 15 minutes? Let's put a timer of 15 minutes. 03:10:10 Rajashekar G: 15 minutes. Let's figure out what can be achieved or 10 minutes. However, whatever you decide the time 10 or 15 minutes. How do I stitch the gap between you and Yeshuan Wenger? What are the right questions to be asked? Venkatesh Tammareddy: first even before if I want to ask questions I need to understand what we are trying to solve. Rajashekar G: Okay. uh Venkatesh Tammareddy: Yeah. Rajashekar G: NLP Yeshwanth Reddy Yerraguntla: Good. Venkatesh Tammareddy: Yeah. Rajashekar G: GP you ask a question a question based you show some some Venkatesh Tammareddy: Yeah. Rajashekar G: information okay so and we are only right now the for the scope of this discussion we are focusing only on data visualization swing Okay. Venkatesh Tammareddy: Okay. Rajashekar G: So quy some data it will throw usually as data visualization. Okay. Data visualizations that maximum 16 most popular are top six graphs column line pi. Okay. Venkatesh Tammareddy: Okay. Rajashekar G: They stuck to those six graphs. Venkatesh Tammareddy: Okay. Rajashekar G: Okay. Okay. Venkatesh Tammareddy: Okay. Rajashekar G: Now this is what is built on enterprise brain. 03:12:02 Rajashekar G: But this is not enough. Venkatesh Tammareddy: Okay. Rajashekar G: I mean the the patterns we have to use have to be very very different because of the volume of the data and the type of data that a models have to consume. The whole point NLP because one is convenience. Second thing especially when AI is involved. Okay. Now visualizations now go to that Venkatesh Tammareddy: One. Rajashekar G: flourish already data visualization examples. Okay. Venkatesh Tammareddy: Yeah. Rajashekar G: Interactction multiple Venkatesh Tammareddy: Yeah. Rajashekar G: diagnostics. But we may have if we may have to we may have to replace that kind of a flow with one of the graphs that I'm showing there right now on flourish. Okay, this is one side of the aspect. Venkatesh Tammareddy: Okay. Rajashekar G: The second side of it is that's what I am understanding right now and what is saying is which graph to Venkatesh Tammareddy: Okay. Rajashekar G: pick. Okay. Venkatesh Tammareddy: Yes. Rajashekar G: Now, do you guys have the capability to do the front end experiential design and front end engineering is nothing but 03:14:17 Venkatesh Tammareddy: Yeah. Rajashekar G: graphs. That's what they're asking you. Okay. Venkatesh Tammareddy: Okay. Yeah. Rajashekar G: So what questions to ask or how to take this conversation forward? I need you to take over from here. You are ashwan. One of you lead the conversation. Venkatesh Tammareddy: Yeah. Yeshwanth Reddy Yerraguntla: Right. So I can ask a starting question um in uh whatever in front end uh React or JavaScript to come up with these visualizations. So I can understand that these are individual components that can always be customized. Venkatesh Tammareddy: And it depend. Okay. So it depends actually um okay so there are two things Yeshwanth Reddy Yerraguntla: Mhm. Yeah. Venkatesh Tammareddy: right market there are libraries that are available right. Yeshwanth Reddy Yerraguntla: Correct. Correct. Venkatesh Tammareddy: So we need to integrate one of those libraries right but uh uh if you're going with that approach right uh then you are so we have to choose one one library right at this point of time but then otherwise we have to build all these components using some third party libraries like JS and everything but that is an effort by itself actually right M so tools 03:16:00 Yeshwanth Reddy Yerraguntla: Break. Break out there. Venkatesh Tammareddy: like Yeshwanth Reddy Yerraguntla: Wi-Fi. Venkatesh Tammareddy: yeah okay full signal is it better Yeshwanth Reddy Yerraguntla: Yeah, it's better. Venkatesh Tammareddy: now. Yeshwanth Reddy Yerraguntla: I guess if the question is uh should we explore outside libraries or should we build something ourselves? Naturally, we should gravitate towards something that already exists and not reuse sorry not Venkatesh Tammareddy: Yeah. Yeshwanth Reddy Yerraguntla: reinvent. In that case, uh then my follow-up question is going to be like uh Venkatesh Tammareddy: Yes. Yeshwanth Reddy Yerraguntla: um what are the let's say this is an alpha version that we are building right so to mainly to get a buy in from Venkatesh Tammareddy: Okay. Yeshwanth Reddy Yerraguntla: Navin Nana and Pratma that yeah this is this is exactly how I'm envisioning um even though it is very uh basic right now this is in the right section. What are the things that we should be uh able to show on the uh Venkatesh Tammareddy: Yeah. Yeshwanth Reddy Yerraguntla: demo that what kind of questions when we ask what are the different 03:17:08 Rajashekar G: What's this? Yeshwanth Reddy Yerraguntla: types of u visualization uh visualizations that we should be able to show and what is the complexity in each of those visualizations that we should show so that we can get a very proper buy in that's the next question we have to answer. Venkatesh Tammareddy: Understood Ashwan. So Ashwan Dhaniki there are two ways Ashwan like there are opensource platforms like D3.js right so where again on top using D3.js JS you can build your Yeshwanth Reddy Yerraguntla: Got it. Venkatesh Tammareddy: uh uh custom visualizations right that is one Yeshwanth Reddy Yerraguntla: Got it. Venkatesh Tammareddy: approach and the second one is like you can go and uh buy something like high charts or there are other libraries that are available right you pick you pick them and for the first version what we can do is we will say these are the chart types that are supported and when you're asking a question on everything I mean let's say for example ask the question Rajashekar G: You can do this Venkatesh Tammareddy: right and you gave it to the model and to the model also we we I mean there is a possibility to say that 03:18:14 Rajashekar G: one. Venkatesh Tammareddy: these are the current visualizations that I have available please suggest me what is the best suited for this so that is the approach maybe that we can take for the first Yeshwanth Reddy Yerraguntla: So as Rashikar already showed the pylog interface Venkatesh Tammareddy: release Rajashekar G: That's Yeshwanth Reddy Yerraguntla: already supports the basic ones bar line pie chart all those things and the ask is this is not what Venkatesh Tammareddy: Okay. Yeshwanth Reddy Yerraguntla: we want. We want something beyond we which is when Rajar started showing the whole interaction animation that you have in flourish. studio, right? Um so um again um to say that we will support these uh charts or these an these visualizations. I only brought it up but I have a feeling that maybe that is the wrong Venkatesh Tammareddy: Yeah. Yeshwanth Reddy Yerraguntla: uh approach. Venkatesh Tammareddy: So yeah, I don't I have to go through pylog once, right? Let's say current implementation I'm talking about. This is my understanding. When I ask a question, Yeshwanth Reddy Yerraguntla: Yeah, 03:19:41 Venkatesh Tammareddy: you are generating those HTMLs like I mean when I say HTML, those graphs in the Python and you are pushing it to the front end, right? It's not APIdriven basically, right? Yeshwanth Reddy Yerraguntla: it's not. Venkatesh Tammareddy: So everything is generated in everything is generated in the Python code and you are showing it Yeshwanth Reddy Yerraguntla: Yes. Venkatesh Tammareddy: here Yeshwanth Reddy Yerraguntla: So if I want if you want me to go technically I will tell it in 30 seconds. Venkatesh Tammareddy: right? Yeshwanth Reddy Yerraguntla: When you ask a question the agent is deciding the specification of the chart. Venkatesh Tammareddy: Yeah. Yeshwanth Reddy Yerraguntla: It will say I want three charts. First chart x-axis should be this. Yaxis should be that. It should be scatter size should be so and so. Second chart is so and so. Third chart is so and so. Specifications is there is a engine in Python that consumes those specifications and generates the charts. those charts the same engine can u generate and convert them into a HTML uh uh format and this pylog is doing an iframe on that and simply showing the chart. 03:20:40 Venkatesh Tammareddy: chat. Yeshwanth Reddy Yerraguntla: So far that is happening. Venkatesh Tammareddy: Yes, understood. Yeshwanth Reddy Yerraguntla: So the expressity of pylog is depend is constrained by the expressivity of the specification of the supported library. Venkatesh Tammareddy: library in the sense here an agent Yeshwanth Reddy Yerraguntla: Uh-huh. Venkatesh Tammareddy: or what Yeshwanth Reddy Yerraguntla: The rendering library, the chart rendering libraries specification is the bottleneck. Venkatesh Tammareddy: is understood. Yeshwanth Reddy Yerraguntla: Now, Venkatesh Tammareddy: So what do you have any library in Python currently that you're using that Yeshwanth Reddy Yerraguntla: currently the library is called alter a r. Venkatesh Tammareddy: is Yeshwanth Reddy Yerraguntla: It uses what is called as a Vega specification. Rajashekar G: Okay. Venkatesh Tammareddy: okay? Yeshwanth Reddy Yerraguntla: V EJ Vega by Venkatesh Tammareddy: Okay sure so there are two things that we can Yeshwanth Reddy Yerraguntla: itself. Venkatesh Tammareddy: uh do here in this case Ashant right uh one on the Python side I'm Rajashekar G: I mean at 10:00 I go home because it will be till 21:00 11. Venkatesh Tammareddy: not sure how How rich is the visualizations library that we have in Python, right? All uh like the one that you mentioned now uh that we have to explore. 03:21:48 Venkatesh Tammareddy: And the second thing is uh we can also parallelly explore uh the libraries that we are having on the JavaScript side. Right? So in the second case what happens is like your model will tell that you know this is my data and it can suggest this is the graph that I want to render right and to the front end you will send the raw data right and using that raw data in JavaScript we will plot the uh visualization. Yeshwanth Reddy Yerraguntla: Yeah. Venkatesh Tammareddy: So that is uh another way to do it. Yeshwanth Reddy Yerraguntla: So Venkatesh Tammareddy: Basically APIdriven just to summarize an APIdriven uh you know visual generating a visualization basically. Yeshwanth Reddy Yerraguntla: generating visualization or driving through API is not unknown here. What is unknown is how do we come up with the specification that Venkatesh Tammareddy: Mhm. Yeshwanth Reddy Yerraguntla: can uh that is that makes creating those extremely complex visualizations possible not just visualizations but also animations I don't think people are even asking this question or this is a common thing in the world right now if it is common well and good we have to use that. 03:23:09 Yeshwanth Reddy Yerraguntla: So what I will ask you Anna is to explore uh those libraries which support these things like when you see um this flourish.studio studio they are they're hard coding for the use case they say you come to us you tell your problem I will build the story I will build the visualizations done but to generate these visualizations on the fly by AI AI will always say this is my intention Venkatesh Tammareddy: Yeah. Yeshwanth Reddy Yerraguntla: like AI will say I want this two bubble charts first bubble chart should be connected to second bubble chart in so and so fashion it can always give in the language of a human being. But to convert that into a proper specification, an edi uh we have to um really think about it. Venkatesh Tammareddy: And the same. Okay. So, okay. Got it. Ashan. Yeshwanth Reddy Yerraguntla: Yeah. Venkatesh Tammareddy: So, two questions. Ashwan. So, when you say that you know the agent will give in plain English, right? which is understandable by the human. Can we can we can can it also have a does it also have a flexibility to give it in a JSON 03:24:31 Yeshwanth Reddy Yerraguntla: It does. It does prove Vega specification that we are able to generate the charts that you saw on the Venkatesh Tammareddy: format. Yeshwanth Reddy Yerraguntla: screen. When I said plain English, Venkatesh Tammareddy: Okay. Yeshwanth Reddy Yerraguntla: I meant it internally knows how to translate that intention into that JSON or whatever specification format. Venkatesh Tammareddy: Okay. Yeshwanth Reddy Yerraguntla: That is the power of the LLM we want to use. All right. So maybe I was wrong in saying it gives in plain English but what I was trying to say is it understands plain English and gives it in the specification syntax that you um ask it. Venkatesh Tammareddy: Okay. So, yeah. So now what are the and see again I'm confused here Ashant when I say I'm confused right what Yeshwanth Reddy Yerraguntla: Okay. Action action item see go through the Venkatesh Tammareddy: I'm Yeshwanth Reddy Yerraguntla: famous libraries that are out there. See which of those first of all support the complexity that is given by flourish. 03:25:35 Yeshwanth Reddy Yerraguntla: Studio first thing are like okay these these two frameworks actually can do this level of complexity. Venkatesh Tammareddy: Thanks. Rajashekar G: Goodness. Yeshwanth Reddy Yerraguntla: Now if that is possible we have to understand the uh Rajashekar G: Hope this one Yeshwanth Reddy Yerraguntla: internal documentation of those libraries as to how they are generating such complex visualizations via code. Then we have to go one level lower and say okay this is how they are this library is doing. How can I make reusable API components so that LLM can access them and generate those the same specification diagrams that will be the action it Venkatesh Tammareddy: Okay. So first part clear rash one see first partic high level and put it and I I have an answer to that Yeshwanth Reddy Yerraguntla: Okay. Venkatesh Tammareddy: actually right um right you know I mean it's not something that is new or anything like I said right you give me an API like for example Yeshwanth Reddy Yerraguntla: Okay. Venkatesh Tammareddy: your LLM model might have said you know uh you know here uh what do you call a pie chart is a is is a representation that is best suited for this kind of data set Right? 03:26:49 Venkatesh Tammareddy: And when it gives you that hint, you already have a raw data associated to that particular question. Yeshwanth Reddy Yerraguntla: Yeah. Venkatesh Tammareddy: Right? like uh let's say for example uh uh percentage of Android users uh across uh ZOMI whatever it is right you ask that question it says basically you want to share what is the share of each company in Android right now what do you do Rajashekar G: Oh, Yeshwanth Reddy Yerraguntla: Yeah. Venkatesh Tammareddy: you send I mean pie chart is the representation that we want to use here and is the data that I have with me right and you will send the data in a JSON format or Rajashekar G: this is very difficult. Venkatesh Tammareddy: even if don't send it. If you send that raw data to me, uh these are the uh this is the raw data. Rajashekar G: Hello. Venkatesh Tammareddy: What I will do is on the front end side I will have a logic implemented where I have a library to say that you know these are the charts different okay you will tell me this is a pie chart I understand okay now I have to render the pie chart I'll call my pie chart component and the feed the data that is that you are going to give it to me right at a high level I'm talking about that is how it is going to work in the JavaScript world right Now on the other side of the things what you are also saying is now let's say 03:28:06 Venkatesh Tammareddy: from here to there if I want to do a drill down right the representation might change it might change into some other thing right that is also a followup that that we have to see Yeshwanth Reddy Yerraguntla: Got it. Venkatesh Tammareddy: that is also in another API call right if you ask me to the outside it is in another API call let's say for example I'm drilling it down on a country right uh I'll send those parameters and again you will send Give me back the data that is that you have and you will also tell me um tell me this is the representation for this for this type of data and so on and so forth it will continue. Uh that is one part. The second part is how do we achieve these visualizations. So first we have to define what visualizations do we need we need ashwant right. Then based on that you know we need to evaluate if there are any other li if there are any libraries that are available well and good we'll we'll get those libraries and we will integrate. 03:29:05 Venkatesh Tammareddy: If you don't have one then what we need to do is that we need to take uh uh an open source framework like D3.js and JS and make it as a reusable component on the UI side. Yeshwanth Reddy Yerraguntla: Again, one thing I don't want you to forget is what whatever are these pie charts, line charts, these basic charts, we already have them. These charts can also do drill down. These charts also have interactivity in the sense that I select something some other chart updates. So Venkatesh Tammareddy: Now point what you're looking for is how do we bring those complex uh uh visualizations into the picture. Right. Right. what are the complex visualizations that you that you want then I will get back to you. Okay, this is Yeshwanth Reddy Yerraguntla: definition either designer to give us or we'll have to Venkatesh Tammareddy: the Yeshwanth Reddy Yerraguntla: say um this is possible for first thing design already gave us some input vaguely by saying uh go to flourish.studio studio website designs. I want such kind of uh animations and designs I want. 03:30:34 Yeshwanth Reddy Yerraguntla: So that is the homework on us to understand why they are complex, how they are complex, what are the uh building blocks in those visualizations and why are they different from the things that we are already able to do. I think that will be our first step. Venkatesh Tammareddy: Understood. Sure. But Dani, it's it's a kind of a project that we have to execute at D1, Yeshwanth Reddy Yerraguntla: project Venkatesh Tammareddy: right? Again, Yeshwanth Reddy Yerraguntla: research. Venkatesh Tammareddy: I mean see it's a jargon, right? You're saying research, I'm saying it as a project because at the end of the day, there is a team that is going to sit and actually do it together. Then only we'll understand whether that is achievable or Yeshwanth Reddy Yerraguntla: Now I will suggest only the leader sit on this right now. Venkatesh Tammareddy: not. Yeshwanth Reddy Yerraguntla: Nobody knows what we are trying to do also. So to call it a project is to say that we know what we are trying to do. That is why I'm saying it's research. 03:31:37 Yeshwanth Reddy Yerraguntla: As few people as possible should be involved in this to understand and crystallize the unknowns. Once we know what to do, yeah, definitely we'll get the buy in. We'll make it a project and all that. Venkatesh Tammareddy: But then time. Yeshwanth Reddy Yerraguntla: Seniors only have to do this. Venkatesh Tammareddy: Okay. But the now the first step that we have to do is basically we need to identify how do we achieve the animations that we have in flourish Yeshwanth Reddy Yerraguntla: Correct. Correct. Yeah, Venkatesh Tammareddy: right. Yeshwanth Reddy Yerraguntla: that's the next uh important step. time. I I know everyone will be busy but spending one or one and a half hour should be sufficient in my head because like we are not demonstrating anything other it's just understanding Venkatesh Tammareddy: But Yeshwanth Reddy Yerraguntla: theory Venkatesh Tammareddy: seeing but the problem is not that Yeshwanth Reddy Yerraguntla: time block is the problem Venkatesh Tammareddy: right. No, not not even the time block. Yeshwanth Reddy Yerraguntla: there. Venkatesh Tammareddy: The problem is you know you need to sit and actually at least do one visualization end to 03:32:53 Yeshwanth Reddy Yerraguntla: Okay. Okay. Venkatesh Tammareddy: end then we will understand what is possible and what is not Yeshwanth Reddy Yerraguntla: I see. I see what you mean. Venkatesh Tammareddy: possible. Yeshwanth Reddy Yerraguntla: I know what they for sure. Venkatesh Tammareddy: Okay. So you are saying that the first step that is the first step. Yeshwanth Reddy Yerraguntla: Yeah, obviously then we can analyze and see whether the feasibility really happens or Venkatesh Tammareddy: Sure. Sure. Yeshwanth Reddy Yerraguntla: not Venkatesh Tammareddy: If that is the if if everybody is aligned with your thought process right I'm good with it. But generally right people will go into an assumption saying that you Yeshwanth Reddy Yerraguntla: there is no everybody in this case right now it's just uh the seniors you and I Venkatesh Tammareddy: know we can Yeshwanth Reddy Yerraguntla: will try to work on the technical feasibility of it let's not involve anyone else uh right Venkatesh Tammareddy: sure Yeshwanth Reddy Yerraguntla: now we have enough direction from design um whether we take uh 2 days or Venkatesh Tammareddy: okay Yeshwanth Reddy Yerraguntla: 3 days but max 3 days we'll try to come up with some feasibility report tarata yeah we'll we'll probably take it to navana or 03:33:59 Venkatesh Tammareddy: Sure. Yeshwanth Reddy Yerraguntla: gopalar and see how they respond also Venkatesh Tammareddy: Sure. Yeshwanth Reddy Yerraguntla: yeah meanwhile uh I think wenatana you can drop off Venkatesh Tammareddy: Yeah, I'm good with it. Yeshwanth Reddy Yerraguntla: prair meanwhile to give a salesforce demo to NSL seeds. Venkatesh Tammareddy: Yeah. Yeshwanth Reddy Yerraguntla: What do we do? That that is something we still haven't uh decided Rajashekar G: Yeah, Yeshwanth Reddy Yerraguntla: on. Rajashekar G: that's what has not come yet. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: What I was asking Britma that we have started with a complex problem to be solved rather Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: not complex but kind of a data is complex. Yeshwanth Reddy Yerraguntla: It's just unknown. Rajashekar G: How do we interpret that? Yeshwanth Reddy Yerraguntla: Nobody knows what is the right solution here. Rajashekar G: Yeah. But what I was also expecting is for which uh RMA said I need to reach out to Rakkesh is how do we present the basic UI uh design architecture because Yeshwanth Reddy Yerraguntla: Yeah, if they can give the UI components name modify to polish it, 03:34:58 Rajashekar G: Yeah. Yeshwanth Reddy Yerraguntla: that is not a probably twoour Rajashekar G: See you know Yeshwanth Reddy Yerraguntla: task. Rajashekar G: Salesforce. We have to solve huge managers. Yeshwanth Reddy Yerraguntla: Mac. Rajashekar G: Uh so okay. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: So one while you explore the possibilities of the current the new library for implementing the UI or UI patterns whatever I'll also reach out to Rakkesh to give me the basic design architecture on the the front facing of the enterprise brain necessary Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: bases Yeshwanth Reddy Yerraguntla: Yes. Rajashekar G: engaged in that aspect aspect Salesforce integration context man I'm sorry not context um intent management business uh learning asking for questions system prompt Yeshwanth Reddy Yerraguntla: Right. Rajashekar G: stabiliz we are not there yet. So that'll be Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: good. Yeshwanth Reddy Yerraguntla: Problems I want to Rajashekar G: Yeah. Yeshwanth Reddy Yerraguntla: understand. Rajashekar G: Problems and also listed down related to governance Yeshwanth Reddy Yerraguntla: Okay. Okay. Rajashekar G: but most of the part those kind of things are happening. 03:37:36 Yeshwanth Reddy Yerraguntla: Mark Rajashekar G: Yeah. EB observations Yeshwanth Reddy Yerraguntla: already. Rajashekar G: Uh just a Yeshwanth Reddy Yerraguntla: MD file. Rajashekar G: dark Yeshwanth Reddy Yerraguntla: by you write an MD file, you push it to the correct repo branch, it should show Rajashekar G: as of now. Too much representation Yeshwanth Reddy Yerraguntla: up Rajashekar G: actually. Yeshwanth Reddy Yerraguntla: Just Rajashekar G: Sure. Yeshwanth Reddy Yerraguntla: Mark down Rajashekar G: Sure. Got it. Got it. Yeshwanth Reddy Yerraguntla: and enterprise access. Rajashekar G: Yeah. Yeshwanth Reddy Yerraguntla: So if really he made some progress and he's showing his problems, it show up. It shows up there. Sorry problems. Can I reproduce them in ganch so and so? Can I reproduce it? Rajashekar G: Say your two discuss update. Yeshwanth Reddy Yerraguntla: End of the day whatever they did if it is in GitHub both document I can reproduce it basic not have anything Rajashekar G: Yeah. Yeshwanth Reddy Yerraguntla: to show. Rajashekar G: Whatever the prompt he has changed, I'll share the branch to 03:39:52 Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: you. Yeshwanth Reddy Yerraguntla: S Rajashekar G: 6:00 I'm sharing the branch. Yeshwanth Reddy Yerraguntla: is Rajashekar G: Yes, this one. Yeshwanth Reddy Yerraguntla: going Rajashekar G: So, okay, we have to finalize on the text uh until last time. Yeshwanth Reddy Yerraguntla: He said something. Rajashekar G: say interface news which is kind of say JavaScript based or react based front end there we were able to show something but we had some other challenges then we moved to Pog we were able to make some additional Yeshwanth Reddy Yerraguntla: Right. Right. Rajashekar G: progress we again are switching to a new technology or say whatever we have flourished Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: like application unless it is frozen but at least we'll add more capabilities to it. Yeshwanth Reddy Yerraguntla: to understand the complexities and take the right decisions because react control I had to build this and show and I got separate feedback from both optin team and enterprise team that they were able to do faster work with this Rajashekar G: possible is let Yeshwanth Reddy Yerraguntla: Right. Rajashekar G: us one that's 03:42:05 Yeshwanth Reddy Yerraguntla: Yeah. Uh what is that one thing whether we should continue with pog Rajashekar G: I Yeshwanth Reddy Yerraguntla: or no no this is this is just a hack react is the right way to go forward and Rajashekar G: think Yeshwanth Reddy Yerraguntla: decision either goals or nav. Rajashekar G: we can we do it this way like whichever the team is comfortable with. Yeshwanth Reddy Yerraguntla: Oh, Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: okay. Rajashekar G: Let us build features. branch you'll do research what is possible feasibility study Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: separate branch but the capabilities will be common to both only every time what will happen whenever we want to build everything in the react major efforts wise But it will be more easier. They can make the debugging and changes easily. Right. Yeah. Yeah. We can take a decision. See currently whatever we are doing with Salesforce. Yes. Okay. Opportunity information or leads information or something that I wanted to query. It should be able to present me in a more appropriate format. 03:43:53 Rajashekar G: Graphs beautiful point 3D visualization. So secondary flourish we hit a road then we wanted to revert and then explore something else. Okay. Okay. So stage wise we can put a line that can be you or somebody research every day they'll they'll update that okay this is where I'm I'm done this is where I hit the roadblock this is where it is not possible it is possible this is feasibility study got it So finally what is what goes to production that research decides research you may pull back on the technology you may add new thing that I'm okay and that will lay the foundation for all future applications. Thorough study Yeshwanth Reddy Yerraguntla: correct. But still the decision on which is the mainstream branch which is the research branch clarity for all practical purposes for this one week we'll continue with pilog because developer speed and let's continue with that way anyway if we are demoing it only to navana and other leaders you can always say that The front end is this is just a temporary placeholder because second thing as design right so that is already a blocker. 03:46:06 Yeshwanth Reddy Yerraguntla: So let's not have this discussion of how to uh which one which which lane to take right now because that needs inputs from design inputs from leadership until then let's let them just continue with dialogue anyway it's delivering whatever it needs to Rajashekar G: That's fine. Basically, I wanted to demonstrate the capabilities of the application which we have built so far. Yeshwanth Reddy Yerraguntla: Cut Rajashekar G: The governor is getting delayed already. So and every time there is an issue we are saying that the system prompt needs to be more thorough. Yeshwanth Reddy Yerraguntla: system. Rajashekar G: So how can we claim that we are Yeshwanth Reddy Yerraguntla: How to Rajashekar G: done at least in 80% accuracy 90% Yeshwanth Reddy Yerraguntla: Yeah, Rajashekar G: accuracy? Let us stop there. It's not attain 100% perfection. Yeshwanth Reddy Yerraguntla: it's already except for governance the responses are coming as expected only right Rajashekar G: It's like understand if I exactly say get Yeshwanth Reddy Yerraguntla: still unusable. Rajashekar G: my emails from the last seven days if I say something like so 03:47:25 Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: basically SQL query it's working fine Gmail are good But rag Yeshwanth Reddy Yerraguntla: Uh we have to close it in definitely this Rajashekar G: this week. Yeshwanth Reddy Yerraguntla: week. We have to demo it like uh Rajashekar G: Yeah. Yeshwanth Reddy Yerraguntla: yeah Rajashekar G: Escalations. Yeshwanth Reddy Yerraguntla: when something is failing you can always download the chat itself. This is the chat. Please check on Rajashekar G: Sure. We'll discuss Yeshwanth Reddy Yerraguntla: yeah if something fails if it fails I have to reproduce it Rajashekar G: You're Yeshwanth Reddy Yerraguntla: reproducing we have to knock it off by this Rajashekar G: right. Yeshwanth Reddy Yerraguntla: week no matter what branch and show chat Rajashekar G: Sure. Yeshwanth Reddy Yerraguntla: minimum. Rajashekar G: Sure. Okay. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: Inco query optimization and token optimization. Yeshwanth Reddy Yerraguntla: Are they are they that is the only that is the only thing that you have to do this we can't know. Rajashekar G: Every customer is asking the Yeshwanth Reddy Yerraguntla: Yes. Rajashekar G: samek Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: whenever we are getting all the Salesforce tables. Yeshwanth Reddy Yerraguntla: Okay. Wherever it's failing, Rajashekar G: So, Yeshwanth Reddy Yerraguntla: it's not working. Give me at least 20 30 chats saying we tried this there. Here it worked. Rajashekar G: okay. Yeshwanth Reddy Yerraguntla: Here it didn't work. Rajashekar G: Sure. Flourish related. Yeshwanth Reddy Yerraguntla: Flourish Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: and Rajashekar G: Sure enough Yeshwanth Reddy Yerraguntla: yeah governance. I don't know why is struggling. Transcription ended after 03:51:29 This editable transcript was computer generated and might contain errors. People can also change the text after it was created.