Mar 5, 2026 Meeting Mar 5, 2026 at 02:02 PST - Transcript 00:00:00 Manisha Gundapuneedi: So what is the like entire expectation is like how user flow will be like uh then for me then I feel like it's a it's already you are defining a flow because so far that's what we are doing in the regular also because I'm defining that user will land here from here user will click he will user will land on the next page and this is the one behavior it's a very deterministic page I'm able to define the journey of the user where it is going to end but now we are telling that it's a nondeterministic vertical to My question is like even though we are saying nondeterministic so there will be a common login screen he will land into the page once he landed I mean common thing there once in the landing page itself there needs to be a dashboard for that particular person I'm not thinking like a dashboard for me that is a challenge if you define a dashboard for me the dashboard so far is like we define the dashboard based on the personas and every lawyer we tell that okay we have a three personas based on the personas and their expectations their needs I'm going to define the dashboard right now we don't know what is the user what he needs and all that everything is very dynamic and we are telling it's a hyperpersonalized system we are building so we won't be 00:01:35 Manisha Gundapuneedi: able to do the all the possible dashboard that are needed right right so based on for me that is the reason our framework has to come because we are telling that this persona this archetype of falling into this persona they are interested to view day briefing as the day starts right this particular archetype of same archetype of a different persona they want to see the bottlenecks first like that's a framework is going to define that what the need to come on thing but if you ask me how to show the bottlenecks I will give a a design representation for that but how you going to combine that together in painting a page is a dynamic thing now again now the dashboard is created assume the debriefing is itself is the thing now how when you click on certain things again it's a new query for the user right because there is nothing screen defined there, right? When you click on it again, it will go to your core agent and core agent will pull the restrictive data and show me the evidence to the user. 00:02:31 Manisha Gundapuneedi: So there also the drill downs are very dynamic. We we are not limiting the drill down has to be only two level or three level. It can be n level. We don't know how deeper that you want to go. Everything is in your query and based on that is going to come. But if I go to the three levels down, how I will go the user interface that is this is the path that you travels. Got So that's another pattern design pattern that I can say because previously we used to have a breadcrumbs now what about the breadcrumbs here if I go to three levels down how to do that that is a design pattern that we can give an independent entity okay to you so like that from our end you can expect a a design system or a design language how to control the navigations okay user will land on some page in order to paint that page what are the individual components that you need and once is going to drill down how you are giving ability to user to get back to the previous view or or in any context of any level that you are how you going to invoke your own agent to give your own personalized query to give the responses. 00:03:34 Manisha Gundapuneedi: So these are the individual patterns we are not defining that this is a uh question that you ask for that question you need to have a three drill downs and possible what are the possible that's why we are not going to define because if I define that there will be a number of permutations combination I have to define right I'm just talking we are not giving any deterministic layer to the developers because that's our goal is like it's a hyperpersonalized system even you and me belongs to the same role working on a same project the my landing dashboard will look different from you or even the same widget you're you looking into the performance of the your report is I'm also looking the performance of my report is but you prefer the visual way of looking into it you will have a visual graph there I am a logical person I need a table representation so the same query but output is different that's the reason we are calling it as a very personalized interface For that if I define the flow right then it will be restricted. 00:04:37 Manisha Gundapuneedi: You need only the structure. What is an algorithm? What is a rule right? So we will start with that uh so to begin with we won't be able to define any pages we will be giving an a design system or a design language how to represent certain nodes in the sense because I'm using a node as a thing is a different metaphor I can say. It can be widget it can be an error state. It can be an insight. It can be a warning. What are the things? How can we represent that? And we also identify 10 possible scenarios. For example, day briefing is a one thing and a bottleneck is another thing and u like that we'll define a 10 to 15 scenarios where we can able to explore that's one thing but when the query comes it can be a single visualization or single table or sometimes it can be a visualization along with the table that combination we also have to see how it is going to come. 00:05:31 Manisha Gundapuneedi: So we'll break into phases here. The first phase is like in a single atomic design systematomic language we'll provide. Along with that we'll also give an a navigation pattern when the user drilled down on from landing page to any other page how the user has to come back to the chat interface that navigation pattern also we will define. That is the second thing. The third one we will then we will work on that is a combination visualizations on top of that how the user going to interact because flourish we already exposed that to you right you have seen already interaction how they going to that can be a base for you but now we will see that if the two visualations come how user going to interact that we we have to do like that we'll break into three phases and we'll try to deliver got it sir so also for the hyperpersonalization to work we need to like first place the user into a particular bucket or archetype. Is there like any like agent which analyzes user behavior to place them in the archetype and evolve over time as the conversation proceeds because first conversation we might not have that much data about the user but as they evolve how like the archetypes can change. 00:06:41 Manisha Gundapuneedi: So in the uh core agent itself we are building the user information database where it will be holding. Okay. So it takes like preference we can because core agent will also have a multiple agents we can assure that okay it has lot of sub agents. So because no matter what logical thinking our regular way of execution only the front end is what is like in what the user sees is what the our user interface layer is going to do. The middleware is where the actual routing of the things is going to happen and the signatures where you define is like in the core agent is going to take that call. Entire load is on the core agent threeaxis and the top is like responsible for only the painting the the bottom is only responsible for serving. Got it. Got it. So like in in our design the threeaxis system would go through the core engine. Okay. Because whatever the framework we are going to come that is going to take care on the core. 00:07:39 Manisha Gundapuneedi: And whatever the design system we are giving that is going to taken care by the user interface. Okay. bottom we don't have any that's data source agent that will be working yeah I'm thinking maybe can we give a screen on how we can do new connectors and all the admin interface admin I mean also while designing latency for a query how long does it take for one query It's work like charg that latency you can those are the patterns we need to use because assuming that there will be latency and we need to engage the user we need to create some hooks while something is preparing and all that. So those are all the things that we are part of our design system because no matter what how fast and efficient system it depends upon internet speed and depends upon the the laptop memory and all that the lag will be there always. Okay. And for enterprise Enter. So this one is already built history. Last 12 months stage wise leads come to me. 00:09:16 Manisha Gundapuneedi: No, Vara Kumar Jagarapu: the Manisha Gundapuneedi: no. Vara Kumar Jagarapu: game. All right. Yeah. Manisha Gundapuneedi: share. I'll share And uh so here is working on design system. You can interact with him. So this is Vara Kumar Jagarapu: Constant Manisha Gundapuneedi: one looking very small window. Yes, he's also asking what kind of So this is what happens like I asked the question and it gave me the answer and this is the representation it is showing. Okay. Now this is like no matter what whatever the interfaces that you see the LLM all look nothing different but we want to change the enterprise brain interface and all that. I think it can be mix of like dashboard and chat interface. This we are not able to eliminate this completely. But how we going to bring this to the user has to be little engaging. Yeah. These are the graphs that you're using. Okay. I think is working on it once done. 00:12:12 Manisha Gundapuneedi: Okay. also exactly components if they're not there generate as per the use case that we need to explore particularly I think that would be very beneficial I think so some use cases component system we don't have the proper representation or the best representation for that particular archetype So if we have guidelines for the agent. Yeah, that is what we doing right. We are just defining asking for that will be connected to LLM knowledge loss to me. So they are just selecting the components and they are sending it. So what I'm asking is if we don't have that component for use GP JSON like part best practices for defining this can the agent dynamically generate for that particular architect based on probably able to but then the styling based on the nature of the data you wanted to dynamically render a certain kind of best fits that yeah they're able to they can able to do like for example earier we have not fed it with sanki diagrams Okay. But when we ask for it explicitly generate a diagram for this scenario because it's an right it can able to generate all the things because the library whatever it has to support that graphic. 00:14:01 Manisha Gundapuneedi: Okay. So we will give the like variables basically like the color tokens and so it matches and it will change it from that. Okay. Okay. like we need to deliver something to next week. Whatever you need, we need the design system that will figure it out. And then interactions interactions based on if you can categorize them then okay at least for version one that is good enough to start okay on top of it if you want to add more and more then we'll do it later Not know exist you can zoom in zoom out but whatever by applying some filters on it. Drag and drop. Okay. So that is kind of I can have a rectangle there. Vara Kumar Jagarapu: The mostly functional which graph supports by default. Manisha Gundapuneedi: Okay. For example, how much customization that your design graph library is doing? Can we define the the width of the columns those things as a guidelines? For example, styling of these graphs. 00:16:28 Manisha Gundapuneedi: Styling of these graphs we can change only. We can like colors. Okay. And status because whatever we are using can what the interactions are experienced what we have seen the flourish can be achieved here or not. Sure. Take one simple thing and can try. Okay. and assume that because there is an in flourish we have an example called an open price story right so they assume that the user ask for this thing this particular question what are the top 10 revenue generating branches throughout the world or country okay so we know that once we know that how we ask the next question assume that potentially identifying what are the three possible questions the user can ask immediately so giving that the Nobel prize example we can replicate here. Okay. You and see it. Sure. Okay. The story also we we are able to predict what he might be interested in automatically the story to him. Okay. visible. What comes next? What is the 00:18:00 Vara Kumar Jagarapu: Thank you. Manisha Gundapuneedi: future? Do you have any questions on this particular related to multi-dimensional? Anything else? Vara Kumar Jagarapu: I just want to understand the simulation you have whatever you Manisha Gundapuneedi: Okay. Vara Kumar Jagarapu: guys because like I just want to understand uh what is possible Manisha Gundapuneedi: So reach with your question Vara Kumar Jagarapu: after uh we are getting that atomic designs or Manisha Gundapuneedi: is Vara Kumar Jagarapu: user navigation whatever it may be If you can understand how we can accomplish this then maybe I can give better Manisha Gundapuneedi: okay. Vara Kumar Jagarapu: ideas. Manisha Gundapuneedi: So can you connect after that? Okay. foreign laptop. I'm sorry. Keep Vara Kumar Jagarapu: Maybe better. Okay. Manisha Gundapuneedi: on mute. That was good. Vara Kumar Jagarapu: Canist. Manisha Gundapuneedi: So basically we defined multiple personas for multiple users and we define UI layers also. I mean what kind of uh UI charts that user is interested in actually. Vara Kumar Jagarapu: Okay. Manisha Gundapuneedi: I mean if if the user person is controller tuner he might be interested in charts he might be interested in tables and all but that we defined already we defined that and there is some data. 00:22:43 Manisha Gundapuneedi: So agent is analyzing the data and agent is deciding which UI to I mean render in the front end actually. So it is giving a JSON using that JSON we are rendering in the front end. Right now it said this card this table this table actually and this supporting evidence this is also kind of card actually and these cards it suggested these things will be interested I mean these things will be useful for that particular person if we change the persona then this dashboard will change again can you scroll to the top on select current based on the executive profile you're creating this dashboard dynamically and change profile. What is this change profile? Change. Vara Kumar Jagarapu: Come Manisha Gundapuneedi: This is just for I mean the data which we have in the Vara Kumar Jagarapu: on. Manisha Gundapuneedi: sheet. Vara Kumar Jagarapu: Get away. Manisha Gundapuneedi: Okay. Fine. Okay. Now this same whatever we have done. So right now so now based on that profile you are thinking that this data is needed for him. 00:24:09 Manisha Gundapuneedi: So right now you consider only the archetype of user and the archetype of the persona. art but we have another three sheets there the data and the possible representations of the model also consume model is also there okay now this particular data falls under which design data archetype uh this is module prediction without prompt so in that sheet actually we have mentioned let's see I'm just trying to understand how the linking has happened uh we mapped it in the back end Clean. Okay. Okay. Fine. Yeah. So this is the US schema that we mapped. Let's say uh if it is a daily briefing for that daily briefing uh we are telling a to generate any of these uh cards KPI tails or bar chart actions evidence any of these things. Okay. Can you open the Excel sheet also? This is uh you are referring to the one last sheet of it. actual Excel sheet architect. Okay. Architect is different. Yeah. Last sheet. 00:26:38 Manisha Gundapuneedi: time series comes to the system Your representation has to be these things right. So each level already. Daily briefing is a combination of multiple things. system. So next oneapility. Okay. Who's that? system component on that front end we already defined the styles for the table charts and everything. So based on that we are rendering back end is only just sending the data. Yeah, got it. Manisha and we discussed it earlier. If we give an atomic things, now the system has to look into those atomic design elements and has to create the page. Correct. Okay. We haven't mapped all the roles actually. Revenue probability. So multiple system we deliver this design system we'll try to define some inspirational designs to you which can able to we can aim even though it's generated system generated dynamic pages but what is the difference that we are expecting that references we can able to provide how much closure you can able to come to that You can see. 00:30:53 Manisha Gundapuneedi: Okay. The third one is like interactive layers because multiple graphs in a single view most of those things cover in the expression itself. How two graphs comes with interaction layers and all that but ultimately right whether you use any other design charting tool we don't have any issues any library. Okay. But we are telling that in the smoothness and experience has to be closer to the flourish. Good. So flourish has this feature where there are transitions from one shot to another shot. Right? So that is the like main point of this because like it can tell a story through that visual system. Yeah, we can use flourishing and in the flourish state again it's like a static only someone already that these are all the seven things that user can do they already defined it. So already for all the seven queries that's going to come this data is already there. So based on that that's the reason that even though you're switching the context it looks like a very smooth rendering. 00:32:03 Manisha Gundapuneedi: It's like a movie animated thing. In our case, if you're expecting the user to go and talk to the GPT and get that the response will have something. So our response smoothness of rendering but still when the comes how we can able to transform that is another thing that we can look at if you are able to predict the possible scenarios assume that the data is also with them along with the possible two questions we know that 10 will be 10 will be next month. So in the same way when we know the possible two potential questions. Okay. Okay. This is good. And we understood and we also set the expectations right what you can expect from us will work. Okay. Vara Kumar Jagarapu: component. Example, user navigation Manisha Gundapuneedi: Uh for example, Vara Kumar Jagarapu: pattern. Manisha Gundapuneedi: the depth the user can go is three levels or four levels. Assume the dashboard. We can able to define the path of the journey. Vara Kumar Jagarapu: H. 00:33:53 Manisha Gundapuneedi: every time you can ask to come one step back. Vara Kumar Jagarapu: Yeah. Manisha Gundapuneedi: How we are going to establish that pattern? That is one thing. The Vara Kumar Jagarapu: Okay. Manisha Gundapuneedi: second Vara Kumar Jagarapu: Okay. UI when you have components browser back button that is one way. Manisha Gundapuneedi: units when we give some inspirations to Vara Kumar Jagarapu: Um Manisha Gundapuneedi: you because we define pages to Vara Kumar Jagarapu: okay. Got it. Manisha Gundapuneedi: give the Vara Kumar Jagarapu: Got it. Okay then. Fine. Fine. Manisha Gundapuneedi: Yeah. Vara Kumar Jagarapu: Okay. Manisha Gundapuneedi: Any other Vara Kumar Jagarapu: No, no, nothing. Manisha Gundapuneedi: questions? Vara Kumar Jagarapu: Okay. Just a pattern sentence and I didn't expect that we'll be giving some. Okay. Manisha Gundapuneedi: Yeah. If you give some differences maybe first agent user def always the navigation and contextual has to be this Left Vara Kumar Jagarapu: design Manisha Gundapuneedi: side. Vara Kumar Jagarapu: system. Okay, for now we are going with one. But Manisha Gundapuneedi: Okay. Vara Kumar Jagarapu: yeah. Manisha Gundapuneedi: So the Vara Kumar Jagarapu: Um Manisha Gundapuneedi: styling Vara Kumar Jagarapu: yeah. Manisha Gundapuneedi: I'm sharing example actually these are the colors we should follow text which colors primary secondary datab. Vara Kumar Jagarapu: Yeah. Manisha Gundapuneedi: So likewise we will get things for this architecture how to represent a button because typically any other design system for field so Vara Kumar Jagarapu: Yeah. Manisha Gundapuneedi: system. Vara Kumar Jagarapu: No. Yeah. Yeah. Manisha Gundapuneedi: So styling complete for application we didn't define Vara Kumar Jagarapu: Yeah, Manisha Gundapuneedi: it. Vara Kumar Jagarapu: just colors like we can manipulate but yeah as a middle UI different components those components also we need it Manisha Gundapuneedi: Yes. Vara Kumar Jagarapu: right we can use directly them using programmatically yeah got it connected with that Thank you Transcription ended after 00:37:41 This editable transcript was computer generated and might contain errors. People can also change the text after it was created.