Apr 2, 2026 Enterprise Brain Scrum - Transcript 00:00:00 Yeshwanth Reddy Yerraguntla: Good. Rajashekar G: Hi. Good Yeshwanth Reddy Yerraguntla: Good morning. Rajashekar G: morning. And I have seen your points on Ablash document. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: Sorry, we'll make sure they are indented. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: H actually and I think they are not happy with the outputs right now. Whatever we are getting on data Yeshwanth Reddy Yerraguntla: and water. Rajashekar G: official approved. So I need to go through them. We are storing everything in the database right now. So we'll go through the logs and then we will correct the system from right now we haven't any spoon feed or business logic nothing we have given only it is understanding the system by the schema schema like on the tables names itself it is having some description like what to do which table is what we need we need to improve Yeshwanth Reddy Yerraguntla: Interesting. Rajashekar G: that business logic into the system right Yeshwanth Reddy Yerraguntla: Understood. But was there any qualitative feedback? Not saying uh they didn't like it is very great. 00:03:17 Rajashekar G: Yeah, exactly. Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: I need to see we'll take it on new Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: office and other than this this document was getting ready and Yeshwanth Reddy Yerraguntla: Set. Rajashekar G: UI related we are working with Amulia design branch she's looking into it I'll connect with V also on that and get it ready Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: So whatever the design system is that is going to change actually. Hi bro. Vara Kumar Jagarapu: Hey, Rajashekar G: Bro, uh you were there yesterday like Ninser has given some feedback. Vara Kumar Jagarapu: 7:30 now. Rajashekar G: Okay. Vara Kumar Jagarapu: We need to catch them. Rajashekar G: Yeah. Vara Kumar Jagarapu: Morning 9:30 got some Rajashekar G: Okay. Vara Kumar Jagarapu: band. Rajashekar G: So actually what I have done is a PR ratio. I have taken another branch called Chanika/BA. So in which uh there is only uh RT agent and uh this uh Tata steel data source agent only two ports only we are making up and all the UI part and all the back end part implementation. 00:05:29 Rajashekar G: So I have a PR to chica/ alpha. Yeshwanth Reddy Yerraguntla: All Rajashekar G: We not right away merging it but we need to verify Vara Kumar Jagarapu: Okay. Rajashekar G: everything. Vara Kumar Jagarapu: Boiler. Boiler plate. Rajashekar G: Yes. Vara Kumar Jagarapu: actually wanted them quality. Yeshwanth Reddy Yerraguntla: right. Rajashekar G: Right. Vara Kumar Jagarapu: Yeah. Rajashekar G: Get some points. It will get much better. Vara Kumar Jagarapu: Okay. So call Rajashekar G: Okay. Vara Kumar Jagarapu: evening most Rajashekar G: Okay. Okay. Now, Vara Kumar Jagarapu: discussions. Rajashekar G: okay. Evaluation related now. How we'll start like uh yesterday Brahakasar is uh trying to connect the system with promptu or something on Jap. Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: So that is one thing and another one coming to the test uh Yeshwanth Reddy Yerraguntla: All Rajashekar G: I mean code unit tests is writing for the front end unit test and for back end also we need to write all the test files and run them that is one part code evaluation second part is agent 00:07:06 Yeshwanth Reddy Yerraguntla: right. Rajashekar G: evaluation we need to have a mechanism to do the yeah that we need to figure it I'll connect with Praagers today and Yeshwanth Reddy Yerraguntla: Yeah. Vara Kumar Jagarapu: chip in the okay sorry what niner is mentioned let's say that we Rajashekar G: let's Vara Kumar Jagarapu: have the same purpose agents uh three agents or four agents right out of them like given a query how they are answering if there any contradiction between those answers then we can conclude something like there is something wrong uh from one of the agent Right? uh they taking an example for example uh uh the customers right so they're not worried about if they didn't get some I mean complete answer but they're worried about a wrong answer right in that case what he mentioned for if that something agent is related to some medical thing so that should be very accurate at that point right so if you want to test Rajashekar G: Oh Vara Kumar Jagarapu: Right. We need to ensure that uh I mean what we can say reliability. 00:08:26 Vara Kumar Jagarapu: So multiple agents multiple Rajashekar G: s***. Vara Kumar Jagarapu: agents we should get what is intended for exact number differently. If there is something very critical answer we can say like our agent is proper like Okay. Yeshwanth Reddy Yerraguntla: Huh? Vara Kumar Jagarapu: correct Yeshwanth Reddy Yerraguntla: Correct. Vara Kumar Jagarapu: answers. Yeshwanth Reddy Yerraguntla: Fair enough. False positives. We should. Where are they taking the agents from? Rajashekar G: Not the agents actual agents. It's like we are asking the same question to three Vara Kumar Jagarapu: LM Rajashekar G: LLMs. Yeshwanth Reddy Yerraguntla: Fine. Vara Kumar Jagarapu: if three LM is giving same answer. Yeshwanth Reddy Yerraguntla: Yeah. Vara Kumar Jagarapu: So the can say the first one we can Yeshwanth Reddy Yerraguntla: Before going forward, Vara Kumar Jagarapu: consider. Yeshwanth Reddy Yerraguntla: can you give me an example question? Vara Kumar Jagarapu: Okay. One question is uh five uh visiting places in Hyderabad. Let's Yeshwanth Reddy Yerraguntla: Okay. You're talking only general question. Vara Kumar Jagarapu: say 00:10:25 Yeshwanth Reddy Yerraguntla: So you're talking you're explaining the concept that three LMS when they agree that means uh the okay the underlying answer is Rajashekar G: more confidence, Vara Kumar Jagarapu: H Yeshwanth Reddy Yerraguntla: correct Rajashekar G: right? Yeshwanth Reddy Yerraguntla: that part is There. Vara Kumar Jagarapu: Yeah, I mean he just given uh lecture but yeah in this just I'm giving that Rajashekar G: future help. Vara Kumar Jagarapu: context Rajashekar G: We Yeshwanth Reddy Yerraguntla: Consistency evaluation is a common paradigm in LLM evaluations that I agree Rajashekar G: going Yeshwanth Reddy Yerraguntla: with. Oh, the the major unknown here is what are the questions we are asking. So Rajika irrespective of what testing method we take up right I would say don't even worry about which Rajashekar G: All Yeshwanth Reddy Yerraguntla: method focus only on which uh questions are we going to ask and how it should answer that is critical once Rajashekar G: right. All right. Yeshwanth Reddy Yerraguntla: you have the data set uh asking questions evaluating is literally two three hours job focus only on Rajashekar G: Yeah. I'm I'm also more worried about the questions only. 00:11:40 Rajashekar G: It's only like previously I got the questions from solve it but right now uh we tried with those questions those questions are giving right answers we have deployed right now the intention might be vary for Navincer and Praakasar based on the business Yeshwanth Reddy Yerraguntla: Oh, see you don't have to worry in that case that you did a bad job or whatever we deployed is bad. Rajashekar G: logic Yeshwanth Reddy Yerraguntla: You you evaluated it and it is working. It's just that what we evaluated against is not what they were expecting. Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: So get a sign off from Prabhakar with the same questions. Simple. We'll escalate this saying this is what we tested on. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: This is what was the report. If there is a gap then the gap is uh in our understanding not in the Rajashekar G: Okay. And one one more thing in the data itself there are some gaps actually I have seen lot of Yeshwanth Reddy Yerraguntla: execution. Rajashekar G: columns might be null in the data. 00:12:34 Rajashekar G: So based on that also it is uh doing like that. That also we need to look into Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: today that I'll discuss with uh them and then I'll take a call where to Yeshwanth Reddy Yerraguntla: What apps are there? Rajashekar G: improve. Yeshwanth Reddy Yerraguntla: I didn't uh Rajashekar G: I'll share my screen. Vara Kumar Jagarapu: Goodbye. Rajashekar G: So let's uh Vara Kumar Jagarapu: Sorry. Rajashekar G: things. Yesterday I have seen one. So while testing let's say uh while pushing this this data into the table value is like same as P above only. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: So one on the first iteration we haven't seen this. Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: So this data this row particular data this has come as a null value while testing found out something is like this. Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: So this data this value need to be here. So we kept it in the DV we wrote a query and updated the value. Likewise. Yeshwanth Reddy Yerraguntla: When you say we kept it in the DB, 00:14:14 Rajashekar G: Uh Yeshwanth Reddy Yerraguntla: uh are you saying you actually modified the customer data or is this DB for evaluations Rajashekar G: we haven't modified this in the sheet. We have modified it in the database like we needed. Yeshwanth Reddy Yerraguntla: then? Rajashekar G: Why? Because here this is human readable. Whenever I ask a question about this particular row, Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: the answer in my database is null. Why? Because this is a Yeshwanth Reddy Yerraguntla: Got it. Rajashekar G: string Yeshwanth Reddy Yerraguntla: So, okay. Those you can change. Okay. This is data normalization. Rajashekar G: data nomination. Yeah, this kind of point still might be there. Yeshwanth Reddy Yerraguntla: I see. Rajashekar G: We need to see where are all those gaps. Still we need to figure it out. Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: And yesterday Pan has started working on this uh uh risk management data. So we connected with Gopaler and what he suggested is first of all put this data in dump into an SQL I mean database first then you can find the relations first and Japer. 00:15:24 Rajashekar G: So we are following that approach right now. There are a lot of sheets. Yeshwanth Reddy Yerraguntla: Okay, Rajashekar G: So even we wanted to find one relation from I think it is taking lot of Vara Kumar Jagarapu: Good night. Rajashekar G: time manually if you do through the Yeshwanth Reddy Yerraguntla: but you can ask RGP or Claude. No. Rajashekar G: he tried with solve it or not still it is taking some Vara Kumar Jagarapu: Happy Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: time. So he said like SQL is a better place to understand like first of all dump everything into SQL then write queries faster in Yeshwanth Reddy Yerraguntla: Mhm. Rajashekar G: Japan Yeshwanth Reddy Yerraguntla: I don't really understand. You can write uh same panda squares in solve it also. Rajashekar G: and Pan said he he might connect with you today if you need some inputs. Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: Why? Because uh in one row there are like uh till by columns let's say let's say there are 20 columns in this sheet in another sheet not 20 those are 30 50 likewise it is there in another 00:16:35 Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: folder Vara Kumar Jagarapu: Accept. Rajashekar G: status whatever he has done yesterday. We need to sync up once then we need to go Yeshwanth Reddy Yerraguntla: Heat. Rajashekar G: further. Sure. Vara Kumar Jagarapu: What's happening? Yeshwanth Reddy Yerraguntla: Sorry. I'll wait for uh power to get back there. Rajashekar G: Uh till what time you will be available or not? Today Yeshwanth Reddy Yerraguntla: Uh 11:30 Rajashekar G: sure he's on the way to office and once he is there I'll ask him. Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: So in between you'll be having genrex calls also right? Yeshwanth Reddy Yerraguntla: Yeah. Yeah. Next I have Rajashekar G: Yeah, I'll ask him Yeshwanth Reddy Yerraguntla: genrex. Vara Kumar Jagarapu: I think from 10 to 10 you schedule another call, Rajashekar G: to Vara Kumar Jagarapu: right? Yeshwanth Reddy Yerraguntla: Yeah. Yeah. Rajashekar G: One is den what I call tenant Vara Kumar Jagarapu: H 9:30 Rajashekar G: better. Okay. So, can we make it online like Vara Kumar Jagarapu: office. Hello Rajashekar G: that's better? Okay. Okay. Sure. Okay. Okay. Transcription ended after 00:19:29 This editable transcript was computer generated and might contain errors. People can also change the text after it was created.