Enterprise Brain Scrum - 2026/03/17 19:56 PDT - Notes by Gemini Mar 17, 2026 Enterprise Brain Scrum - Transcript 00:00:00   Vara Kumar Jagarapu: Sure. Yeshwanth Reddy Yerraguntla: Hi. Vara Kumar Jagarapu: Yes. Yeshwanth Reddy Yerraguntla: fire. Let me Vara Kumar Jagarapu: Access. Yeshwanth Reddy Yerraguntla: check. Vara Kumar Jagarapu: Let's Yeshwanth Reddy Yerraguntla: Yeah. Yeah. That's true. Vara Kumar Jagarapu: be Manisha Gundapuneedi: Add on the bit. Yeshwanth Reddy Yerraguntla: Let me share my screen. Tell me if your emails are wrong. It did you Vara Kumar Jagarapu: Yeah, it is correct Yeshwanth Reddy Yerraguntla: log out and login? Rajashekar G: Ganesh mail is wrong. Yeshwanth Reddy Yerraguntla: I'm checking. Rajashekar G: I'm Yeshwanth Reddy Yerraguntla: See, you don't need to wait on me to give this access. Joshua, Nitish, Rajashekar G: sure Yeshwanth Reddy Yerraguntla: Shraa, they also have Joshua, Vara Kumar Jagarapu: Okay. Yeshwanth Reddy Yerraguntla: Nitish, Shraa So I'm just checking Ganesh Ganesh. Rajashekar G: I'm sending you Vara Kumar Jagarapu: Let me delete this catch and everything. Yeshwanth Reddy Yerraguntla: Yeah. Yeah. Just log out. Login. It should work. Vara Kumar Jagarapu: Yeah, I got it. 00:03:21   Vara Kumar Jagarapu: Listen. Meanwhile, one thing uh so team different places which is not Yeshwanth Reddy Yerraguntla: Okay. Vara Kumar Jagarapu: inv account. Yeshwanth Reddy Yerraguntla: But for which npm are you saying? Vara Kumar Jagarapu: Okay. For example, Yeshwanth Reddy Yerraguntla: Okay. Uh yeah, I think Vara Kumar Jagarapu: yeah, maybe we need to move it. Yeshwanth Reddy Yerraguntla: okay. Vara Kumar Jagarapu: Yeah. Yeshwanth Reddy Yerraguntla: Why will like I just want to understand how this is connected to or is this connected to our Vara Kumar Jagarapu: Just Yeshwanth Reddy Yerraguntla: discussion Vara Kumar Jagarapu: meanwhile. Yeah. make personal ping. Uh so we are maintaining all these reports here actually npm Yeshwanth Reddy Yerraguntla: Uh okay. Vara Kumar Jagarapu: packages. Yeshwanth Reddy Yerraguntla: H what do we need to move Vara Kumar Jagarapu: So packages publish those should come Yeshwanth Reddy Yerraguntla: here? Vara Kumar Jagarapu: here. Yeshwanth Reddy Yerraguntla: We used to have the sorry D here. Let me check anyway. Vara Kumar Jagarapu: Sh. Yeshwanth Reddy Yerraguntla: Nobody's using it. And uh 00:05:39   Vara Kumar Jagarapu: Okay. Yeah. Yeshwanth Reddy Yerraguntla: yeah, Vara Kumar Jagarapu: Okay. Yeshwanth Reddy Yerraguntla: yeah, I'll remember What is the name of the current sprint? common component and Rajashekar G: D Yeshwanth Reddy Yerraguntla: size uh Rajashekar G: sprint DB sprint 1 square 1 DB Yeshwanth Reddy Yerraguntla: spirit. I'm okay. Rajashekar G: sprint 1 Back Yeshwanth Reddy Yerraguntla: My bad. I'm looking at the wrong Rajashekar G: to Yeshwanth Reddy Yerraguntla: guy. Sprints. Epic. Rajashekar G: You're there now. Yeshwanth Reddy Yerraguntla: DB289 platform foundational and core agent infrastructure agent intelligence training platform capabilities. What is the difference? Squad on squad 2 are Vara Kumar Jagarapu: VP Rajashekar G: Uh, Vara Kumar Jagarapu: milestones. Yeshwanth Reddy Yerraguntla: A1 I like I don't know how to read these things. Rajashekar G: I'm sorry. Yeshwanth Reddy Yerraguntla: So either Rajikar wants to share or Rajashekar G: Okay, I'll show Yeshwanth Reddy Yerraguntla: filters. Vara Kumar Jagarapu: So command Take Yeshwanth Reddy Yerraguntla: Oh okay okay okay. I can see them. 00:08:08   Vara Kumar Jagarapu: much Yeshwanth Reddy Yerraguntla: Yeah, this is print one score one there should be Rajashekar G: Now we Yeshwanth Reddy Yerraguntla: a way to group by people also. Vara Kumar Jagarapu: Yeah, Rajashekar G: can Vara Kumar Jagarapu: I think there is a group uh Yeshwanth Reddy Yerraguntla: No. Rajashekar G: select. Vara Kumar Jagarapu: option. Can Yeshwanth Reddy Yerraguntla: Okay. Okay. Vara Kumar Jagarapu: you Yeshwanth Reddy Yerraguntla: Apparently squad two is a court team by Rajashekar G: Wow. Vara Kumar Jagarapu: Yes. Yeshwanth Reddy Yerraguntla: okay I did not touch anything since two so I'll try to get something done today mainly I was occupied with uh uh genrx and other things. So today I'll try to spend as much Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: time. Rajashekar G: So let's go one by one. So Abilash has completed this part like knowledge base setup and cut operations. Our knowledge base has been follow. He has worked on this LLD. Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: So here he'll do it this one and what I suggested him is uh take the next part like org snapshot and knowledge graph is also there right so there's 00:09:51   Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: a story for that so I told him to pick that one and complete the LLD and documentation part Yeshwanth Reddy Yerraguntla: Yeah. Okay. Rajashekar G: today and he has come up with some questions uh we need to clear them Yeshwanth Reddy Yerraguntla: But uh Rajashekar G: off. Yeshwanth Reddy Yerraguntla: Okay. Um, we'll go through that. But this is reviewed. This is okay. Rajashekar G: Uh I reviewed yesterday also you reviewed and gave some feedback Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: modifications. Yeshwanth Reddy Yerraguntla: Right. Okay. Rajashekar G: So right now how we are going is like first of all at least this particular story Yeshwanth Reddy Yerraguntla: Resources. Rajashekar G: is to creating a LLD and getting the clarity like let's close this story and uh create implementation stories new one and uh pick those one. Yeshwanth Reddy Yerraguntla: Makes Rajashekar G: Yeah. Abby can we put the put your questions Yeshwanth Reddy Yerraguntla: sense. Rajashekar G: here? Is there any document you have already generated? Abhilash Adunuri: I'm no one. 00:10:47   Abhilash Adunuri: I'm just uh I have noted all the questions there. Rajashekar G: Yeah, these are the questions you have given me. Abhilash Adunuri: Yes. Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: So on snapshot level Yeshwanth Reddy Yerraguntla: Uh how exactly do we need to store it in database? See question one and three I think are the same things. Ka Abhilash Adunuri: Oh, basic Yeshwanth Reddy Yerraguntla: is yeah understood. Rajashekar G: multiple discussions it's like there are multiple sources multiple sources and how Yeshwanth Reddy Yerraguntla: Mhm. Rajashekar G: it will be is it a DB with separated columns are all sources will be in a single field. Yeshwanth Reddy Yerraguntla: Let us work. uh see I'll give you I will not give you clarity in in terms of how Rajashekar G: Clarity Yeshwanth Reddy Yerraguntla: to do it in DB and all those things what I will give you is how to what is the direction to think in what is the purpose of arc snapshot like this I'm asking this Rajashekar G: Yes. Yeshwanth Reddy Yerraguntla: rajar Rajashekar G: To know the status of the current uh current state of the organization in terms of 00:12:17   Yeshwanth Reddy Yerraguntla: correct No, Rajashekar G: like let's say if I take Salesforce where are we standing today, Yeshwanth Reddy Yerraguntla: no, not in terms of Salesforce or Salesforce could be one way of asking Rajashekar G: right? Uh yes, a snapshot means it will be combined of multiple sources Yeshwanth Reddy Yerraguntla: question. Rajashekar G: status, right? Let's say if I take a project where is that project right now where is that project yesterday so that we need to understand by snapshot Yeshwanth Reddy Yerraguntla: Correct. So followup question is what is the simplest way to uh get this uh uh done? Let's say forget about enterprise brain. We we hired a new intern and their uh purpose is only to keep the snapshot up to date. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: they somehow ignoring the illogical statement I'm about to say somehow let's say we gave that intern all our emails all Jira access all everything given that given that he is Rajashekar G: Mhm. Yeshwanth Reddy Yerraguntla: very tirelessly working for 24 hours of the day how best can he store the snapshot let's say you are that How will you do it? 00:13:48   Rajashekar G: I will summarize everything. So let's say here uh Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: whatever the information coming from I will be having different sources. Yeshwanth Reddy Yerraguntla: Correct. Rajashekar G: So let's say if I take Gmail what are all the emails I received on that day and what is the summary of it? Yeshwanth Reddy Yerraguntla: All right. Rajashekar G: I'll store Yeshwanth Reddy Yerraguntla: Will you store summary just as a summary of Gmail or will you store summary at an organ at a project level or at a topic Rajashekar G: Yes. Yeshwanth Reddy Yerraguntla: level it's important that you give the um tomorrow morning when Navenar walks into office it's up to you to give the summary in the most concise manner Rajashekar G: All right. Yeshwanth Reddy Yerraguntla: If you store Gmail all Gmail summary will that be beneficial for Navin sir if if the organization has 100 people and I store only Gmail summary in one big document correct it will not be Rajashekar G: No, it need to combine all the sources. Yeshwanth Reddy Yerraguntla: helpful h in what way will you combine now one way two huh 00:15:02   Rajashekar G: uh let's say uh if the person is a project Yeshwanth Reddy Yerraguntla: problem. Rajashekar G: manager so he need to have project level summaries if a person Yeshwanth Reddy Yerraguntla: Mhm. Rajashekar G: is someone like CEO he need to have multiple sections like what's happening in my sales Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: what's happening in my engineering what's happening in my design likewise in those also there Yeshwanth Reddy Yerraguntla: Right. Rajashekar G: will be subsections if I take engineering there will be multiple projects each project against what's Yeshwanth Reddy Yerraguntla: Yes. Rajashekar G: happening? Yeshwanth Reddy Yerraguntla: Correct. Rajashekar G: So let's say if I take a project in that project also if I receive 100 mails on some day in 100 mails like how many are related to each project also I need to figure it out and pull. Yeshwanth Reddy Yerraguntla: You have to put them in separate buckets. Right? So now you're like going in the right direction that it's not enough that I Rajashekar G: Yes. Yeshwanth Reddy Yerraguntla: simply store all emails in one big text file. I'll maintain separate text files for each project and because you said I will want to store uh for tech lead a different kind of info for lead different kind of l info for nser a different kind of info. 00:16:16   Yeshwanth Reddy Yerraguntla: Now we'll have to ask ourselves is that really efficient way to do it or is it okay if I store project level uh information across all different sources in one go for that day and when the time comes when I have to tell Navin what he needs to listen I'll go through that file I'll cut paste those uh maybe there are 300 lines I'll cut paste 50 lines out of that and give to Navin sir, a different uh you know a different 20 lines to Praakar sir another different 50 lines to pratim is that the right way to do it or do you suggest no no I will I will create different copies of the same information for different people so that I don't even need to worry about cut pasting Rajashekar G: If you make the copies, that will be a duplication. That's not the right Yeshwanth Reddy Yerraguntla: the perfect it's not the right. Rajashekar G: way. Yeshwanth Reddy Yerraguntla: So you would rather store at a project level or a topic level what is happening and on demand you'll decide from the same text file which subset of lines to deliver to each person. 00:17:30   Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: So now coming back to the original question, how do we store the arc snapshot? The simplest way to store it is as text files. One text for one topic. Rajashekar G: Text file means in the database a row that contains large text. You're saying okay Yeshwanth Reddy Yerraguntla: Yeah, it yeah it doesn't matter. You can do that. Look it's unstructured, right? Is there any way we can bring structure into the document? If if you can bring it, maybe we can brainstorm. Rajashekar G: No. Yeshwanth Reddy Yerraguntla: Maybe those that structure can go into columns. Think about it then. So I want information at a day level. I want information at a topic level. So these two I can we can probably call them as two different uh uh two different columns. Every day you will constantly increase the date number uh for the same topic. Uh you'll create a fresh row with a fresh 00:18:37   Rajashekar G: Uh enough for the better understanding. Yeshwanth Reddy Yerraguntla: document. Rajashekar G: Uh what is topic here? What are you Yeshwanth Reddy Yerraguntla: Topic I am vaguely using that word topic because it could be anything. Rajashekar G: suggesting? Yeshwanth Reddy Yerraguntla: It could be a project like enterprise brain. It could be uh some corporate event uh you know some Slytherin activity. It could also mean uh some pipeline for hiring candidates. So hiring a specific candidate could be a topic. If there are 10 hiring pipel things happening parallelly for 10 different people all the 10 are different topics. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: Uh we can also have topics in terms of uh um lead generation not lead generation. Um yeah we can do that like different different uh pipelines will be there for different different companies. So company itself can be treated as a topic because um if we take uh gen HRX that's that we are about to receive. Two weeks back it was it's same Gen HRX but the topics were mainly I mean the concepts in Gen HRX were mainly in the sales direction. 00:19:50   Yeshwanth Reddy Yerraguntla: That said, it is still GHRX that is constantly evolving, right? At someday we that content will suddenly change from pre-sales information to engineering information. Execution kind of information will be there. So, company wise, we can have one topic. As long as two topics do not have anything in common, we can consider them as topics. Rajashekar G: Okay, Yeshwanth Reddy Yerraguntla: Is am I still vague? Because I feel I'm still vague. Rajashekar G: it's a lot but Yeshwanth Reddy Yerraguntla: We can start with simply companies uh uh information as a one topic uh project for the time being because ultimately Navin Sarat they are Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: interested in satisfying customers. So one customer can be one topic for us. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: At least for the purpose of building internal enterprise we have uh hiring pipelines we have execution or engineering pipelines Rajashekar G: Are you sharing your Yeshwanth Reddy Yerraguntla: uh I'm not sharing Rajashekar G: screen? Yeshwanth Reddy Yerraguntla: anything right so you have date you have topic Um anything else? 00:21:18   Yeshwanth Reddy Yerraguntla: Ah and you have the content in text format as to what happened on that day. This for the time being is fine because we want to deliver something. We want to execute and show that it works. Rajashekar G: What Yeshwanth Reddy Yerraguntla: After that we can make the whole u uh database into a graph database Rajashekar G: happened? Yeshwanth Reddy Yerraguntla: where instead of uh creating a fresh text file see 90% of the content might not change from day to day right so in a grab database you will probably delete 10% of the nodes and add 10% of the Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: nodes that that is more efficient uh just that I don't I think we have the bandwidth or the confidence to execute that. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: Did that answer the questions? No. I mean at least if not answering the questions, you should have the direction of how to approach the problem. Rajashekar G: We will answer better. Abhilash Adunuri: So you're basically saying uh to separate columns or something based on project or it's a data source or something and while while retriving uh uh consuming all the columns we'll get a snapshot. 00:22:44   Yeshwanth Reddy Yerraguntla: So separating based on date and project will be sufficient not data source we'll combine Abhilash Adunuri: Yeah. Yeshwanth Reddy Yerraguntla: all data sources for that day into one document. Rajashekar G: Okay. Abhilash Adunuri: Okay. So date and product is Yeshwanth Reddy Yerraguntla: This decision might yeah this decision might change tomorrow Abhilash Adunuri: fine. Yeshwanth Reddy Yerraguntla: where Navin Moser might say no no no I'm still interested in just the Salesforce aspect Vara Kumar Jagarapu: Nice. Yeshwanth Reddy Yerraguntla: of all the projects in one shot in that if that is the case then the way we are currently discussing will not be sufficient because now we have to go through all the text files and get only Salesforce data. Still that is to some extent e to some extent possible just that it might not be efficient. So based on how p how we get the feedback we'll try to change the data structure but right now if all you are asking is I don't want this head day just give me what to do I will say go for the date the project uh name and the updates for that day across all data sources in one big text file. 00:24:01   Yeshwanth Reddy Yerraguntla: What is the fourth one? How exactly is RTA? What is RTA? Rajashekar G: RT agent RTI agent real time decision intelligence agent is sending Yeshwanth Reddy Yerraguntla: Oh, Rajashekar G: the Yeshwanth Reddy Yerraguntla: second and fourth I think are low level. Do should I send a JSON or something like that? Right. Um, Rajashekar G: I can fix it. Yeshwanth Reddy Yerraguntla: yeah. Rajashekar G: Okay, Abhilash Adunuri: Yeah, finally get them. Rajashekar G: we'll come up with something and again take the clarity. Abhilash Adunuri: Yeah. Yeah. Vara Kumar Jagarapu: Uh yes shant snapshot. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: Okay. Vara Kumar Jagarapu: chart. So as we have scheduuler it might run in different times right again like when one scheduleuler uh is related to one data source uh at that particular point the data Yeshwanth Reddy Yerraguntla: Correct. Vara Kumar Jagarapu: source will get the information and then our snapshot also will be updated Yeshwanth Reddy Yerraguntla: H correct. So if if Gmail is every 5 minutes, Vara Kumar Jagarapu: uh 00:25:12   Yeshwanth Reddy Yerraguntla: Arc snapshot will keep updating every 5 minutes. Vara Kumar Jagarapu: yeah so when it comes to our back again Uh there there is actually governance also. Yeshwanth Reddy Yerraguntla: Right. Vara Kumar Jagarapu: Uh so if we combine our snapshot so we should consider uh the roles also right. Yeshwanth Reddy Yerraguntla: Uh but arc snapshot is because at an organization level there's no arbback uh question here only. Vara Kumar Jagarapu: Okay. So organization level information will be there but some of the Rajashekar G: s***. Yeshwanth Reddy Yerraguntla: Yeah. Vara Kumar Jagarapu: information can be accessed by some people. Yeshwanth Reddy Yerraguntla: Mhm. Vara Kumar Jagarapu: It's not something like that organization level uh I mean when I'm thinking of Yeshwanth Reddy Yerraguntla: Rasher go to whimsical. Vara Kumar Jagarapu: a text file right Yeshwanth Reddy Yerraguntla: Yeah, I'll tell you what. I understood the question. Vara Kumar Jagarapu: okay yeah so when I think of some Yeshwanth Reddy Yerraguntla: Continue asking. Yeah. Vara Kumar Jagarapu: text file right uh so if there is some point okay let's say it's about sales okay so the data is there at automation level but yeah 00:26:12   Yeshwanth Reddy Yerraguntla: Yeah. Vara Kumar Jagarapu: when we are uh moving this information user aware is Right? It should take a decision based on the text. Is it able to take that decision and control uh this governance? That is what I'm thinking actually. Yeshwanth Reddy Yerraguntla: So we will not put any filter at the snapshot level. Vara Kumar Jagarapu: Okay. Yeshwanth Reddy Yerraguntla: Let it contain even the most sensitive information let it contain because nobody's Vara Kumar Jagarapu: Let's say okay. Yeshwanth Reddy Yerraguntla: going to touch it. Vara Kumar Jagarapu: Okay. Yeshwanth Reddy Yerraguntla: What we need to uh guard against is when user aware agent is about to send the information to every person there it has to u run one filter for every user. Every time it before it is about to send information it will uh there will be one more Vara Kumar Jagarapu: H. Yeshwanth Reddy Yerraguntla: guardrail agent asking is this does this have any sensitive information based on the user. So we'll probably ask question like uh this is what user aware agent has decided 00:27:16   Vara Kumar Jagarapu: H. Yeshwanth Reddy Yerraguntla: to send to goals. Uh these are the rules for goals. Now based on the content and rules tell me if uh this information is having any sensitive thing. If the agent says okay there is something sensitive we'll immediately block it there. If agent says this is good to go we'll continue and send it. just as an example I'm giving like exactly how we'll handle it until we come to that point we don't know but the point is at the Vara Kumar Jagarapu: Oh yeah. Yeshwanth Reddy Yerraguntla: purple blocks we'll have the filters but not at the arc snapshot level Vara Kumar Jagarapu: Yeah. Sir actually like I just gone through the commerce life cycle but not in more detail maybe I can more explore and I can discuss better because like what I'm thinking is uh so it's a kind of only one gate as of now Yeshwanth Reddy Yerraguntla: Okay. Vara Kumar Jagarapu: as in our discussion what I feel so how powerful it is always matters because the sensitive information might be leaked right that is what 00:28:33   Yeshwanth Reddy Yerraguntla: Correct. Vara Kumar Jagarapu: concern Yeshwanth Reddy Yerraguntla: No, no. 100%. Uh, you have to Uh we that like at the box number two is where the most uh intelligence and all our um all our uh whatever smartness has to come into picture. Vara Kumar Jagarapu: What I'm thinking is let's say we have this back at multiple uh steps right every place like uh that is a kind of information if we provide. So the security level is getting increased. Yeshwanth Reddy Yerraguntla: H. Vara Kumar Jagarapu: I mean my intention is for example uh when data source right when there also like Yeshwanth Reddy Yerraguntla: Uh-huh. Vara Kumar Jagarapu: we kept our bank while saving itself like we'll be taking care the Yeshwanth Reddy Yerraguntla: Correct. Vara Kumar Jagarapu: uh uh the roles and everything right. Uh so in this case this is a kind of uh system level information also there and we are storing information that is like till this point organization snapshot will be including all Yeshwanth Reddy Yerraguntla: H Vara Kumar Jagarapu: the data without any uh governance 00:29:33   Rajashekar G: Holy Yeshwanth Reddy Yerraguntla: filter. Vara Kumar Jagarapu: right yeah so what I'm thinking if we can keep uh some Yeshwanth Reddy Yerraguntla: Yeah. Vara Kumar Jagarapu: roles related information also uh to the data point what are the points we are Rajashekar G: s***. Vara Kumar Jagarapu: storing there is an additional step but that will help uh to take decision better decision at user Yeshwanth Reddy Yerraguntla: No, I think that is perfect. I you you really gave a uh proper solution to this. Vara Kumar Jagarapu: inter Yeshwanth Reddy Yerraguntla: While you are storing in the text file also store what information you're storing like yeah this is an email this is this contains sensitive or like some financial information that kind of information if we can store 1,000% that will improve the uh reliability 1,000%. Vara Kumar Jagarapu: Yeah. Uh likewise I'm thinking whenever we are storing some information right because like Yeshwanth Reddy Yerraguntla: Yeah. Yeah. Yeah. With every information packet, Vara Kumar Jagarapu: Yeah. Yeshwanth Reddy Yerraguntla: what you're suggesting is also keep track of how sensitive or what sensitive information is in the 00:30:46   Rajashekar G: Hey, Yeshwanth Reddy Yerraguntla: packet. Vara Kumar Jagarapu: H. Yes. Yeshwanth Reddy Yerraguntla: I am completely agreeing with you. Rajashekar G: Okay. Abhilash Adunuri: I know one more. Yeshwanth Reddy Yerraguntla: I think Gopalasar also was suggesting in this direction. I didn't really it didn't really strike strike me when he was speaking but now now I'm clear this is how we should go forward. Vara Kumar Jagarapu: So is working hard maybe. Abhilash Adunuri: Yeah. Uh what I'm going to ask was uh uh we Yeshwanth Reddy Yerraguntla: Mhm. Abhilash Adunuri: we got in a conclusion saying we need to store at a project level across different sources, right? So can I say uh the data will be I mean uh we going to decide a JSON in such a way that RTA agent should be able to send that in particular JSON so that we're going to store in the TV uh Yeshwanth Reddy Yerraguntla: Yeah, repeat that. Abhilash Adunuri: I'm saying uh the RTA agent should able Yeshwanth Reddy Yerraguntla: I Abhilash Adunuri: to send the project level updates right uh is that the Only thing was Yeshwanth Reddy Yerraguntla: correct. 00:32:12   Abhilash Adunuri: uh if it is going to send uh can I tell uh I'm going to consider uh each row for a project based on the date. So when giving that u I will uh put the date filter and across all the projects I'm going to give the snapshot Yeshwanth Reddy Yerraguntla: You can you can go with that. Yeah. Yeah, you can do that. By the way, Ailas, the 340 document has some syntax errors. like Rajer can open 340 and see. Rajashekar G: Uh yeah now this mermaid was not running. I told him I need to update. Yeshwanth Reddy Yerraguntla: Yeah, just wanted to tell him. Rajashekar G: Yeah. Vara Kumar Jagarapu: uh I generally got small actually from the diagram right. Abhilash Adunuri: Yeah, I Yeshwanth Reddy Yerraguntla: Huh? Vara Kumar Jagarapu: So we are talking about organization and uh uh project and data source right. Yeshwanth Reddy Yerraguntla: Yeah, Vara Kumar Jagarapu: So hard or nation having multiple projects a project can have multiple data sources that means like uh where Yeshwanth Reddy Yerraguntla: correct. Vara Kumar Jagarapu: exactly are we differentiating because like when we get some jer update right that is at the project level right if 00:33:45   Yeshwanth Reddy Yerraguntla: Correct. Vara Kumar Jagarapu: something might be a com uh I'm not sure I'm just getting not getting any data source that might be uh under two projects. Do we have something like that? No. Right. Most probably no. Yeshwanth Reddy Yerraguntla: Um, no no you're you're right in asking that like it could there could be a data source where one row Vara Kumar Jagarapu: Design. Yeshwanth Reddy Yerraguntla: is talking about different different pipelines or different projects. In that case simply agent has to duplicate that same information across all the three Vara Kumar Jagarapu: Okay. Yeshwanth Reddy Yerraguntla: projects. Vara Kumar Jagarapu: So uh yeah whenever the user is asking it might it might be from single project or multiple projects also right okay Yeshwanth Reddy Yerraguntla: Correct. Correct. Vara Kumar Jagarapu: so organization snapshot should contain mult let's say if there in multiple organizations also so there should be multiple organizations under multiple projects and under multiple data sources this is how the information should be hierarchal I mean in a structured way if you can store based on that we can retrieve it and give it properly. 00:34:53   Yeshwanth Reddy Yerraguntla: Correct. The the whole goal of organization snapshot is to Vara Kumar Jagarapu: Right? Yeshwanth Reddy Yerraguntla: segregate information into buckets where each bucket has you can almost tell a story with just reading one bucket and content across buckets is independent of each other. That is the only uh key aspect in organization snapshot. Vara Kumar Jagarapu: Okay. So here right uh where we are storing this kind of uh Yeshwanth Reddy Yerraguntla: Right. Vara Kumar Jagarapu: uh information let's say we have let's say they organization okay and okay okay Yeshwanth Reddy Yerraguntla: That's where you'll store it that like world model is where you'll store that now. Vara Kumar Jagarapu: okay so under this project uh why I'm Yeshwanth Reddy Yerraguntla: Yeah. Yeah. Vara Kumar Jagarapu: thinking is let's say we have two data connectors let's for Jera only one data connect one one J is related to one project another the is related to different project both are same data I mean different data sources let's Yeshwanth Reddy Yerraguntla: Got Vara Kumar Jagarapu: say uh while we are connecting adding as 00:35:58   Yeshwanth Reddy Yerraguntla: it. Vara Kumar Jagarapu: a connector right that information we're thinking to store it at world model only but yeah properly getting visualization yeah I will also frame better uh when when we are Yeshwanth Reddy Yerraguntla: Yeah, I mean just think Vara Kumar Jagarapu: adding some connector how we are Yeshwanth Reddy Yerraguntla: through every connector should have its Vara Kumar Jagarapu: registering uh that ids we need Yeshwanth Reddy Yerraguntla: own unique ID. Vara Kumar Jagarapu: to map it in the world model that what we are thinking under the particular project Yeshwanth Reddy Yerraguntla: Yeah. Yeah. Understood. Agreed. Vara Kumar Jagarapu: yeah generally like AR snapshot is there another way except the text which I'm also thinking Yeshwanth Reddy Yerraguntla: Can you repeat that question? Vara Kumar Jagarapu: the okay art snapture without of like going with Yeshwanth Reddy Yerraguntla: I was smart Vara Kumar Jagarapu: the storing as a text file right just Yeshwanth Reddy Yerraguntla: text. Yeah. Vara Kumar Jagarapu: thinking is that is there any other place Yeshwanth Reddy Yerraguntla: I mean uh we can talk with Gopalar here 00:37:12   Vara Kumar Jagarapu: uh Yeshwanth Reddy Yerraguntla: if you think uh it's worth Vara Kumar Jagarapu: Okay. Yeshwanth Reddy Yerraguntla: talking. Yeah, probably because now in we are in the point of taking a crucial decision should we even use Vara Kumar Jagarapu: Yeah. Yeshwanth Reddy Yerraguntla: structured or should we go for unstructured data because unstructured data might help us do quick uh text search. those kinds of things might also help us. Vara Kumar Jagarapu: Okay. Okay. Yeshwanth Reddy Yerraguntla: I can suggest uh okay I let me suggest something to both Abilash and VA here let us only focus on uh instead of doing implementation details for our snapshot let us just list down as many functional requirements and nonfunctional requirements and use cases as we can right how people are going to need organization snap shot if we can understand all the use cases perhaps we can come up with a better Vara Kumar Jagarapu: Yeah, I think like it's most of the core uh Yeshwanth Reddy Yerraguntla: true true I I 100% agree that the entire enterprise brain there are 00:38:21   Rajashekar G: No. Yeshwanth Reddy Yerraguntla: only two really crucial components one is the real time agent and the other one is arc snapshot so it's worth spending more time that is okay we should spend like couple of more days Heat. Heat. Vara Kumar Jagarapu: Yeah, please. I think like uh let's connect around 11:30 after small meeting we have maybe then we can discuss further based on uh different use cases and we can uh also also speak with Gopalaru on his opinion then we can have one more discussion with the correct yeah Rajashekar G: Okay. 11:30 discussion. Vara Kumar Jagarapu: maybe not at schedule but Rajashekar G: We need to schedule a s Vara Kumar Jagarapu: yeah we need to discuss first then we can go for gopal guru and yeah I think like we can change it at any point of although if we keep some text right like we can get some structure later Rajashekar G: Um before that we we will have a discussion like Vara Kumar Jagarapu: on yeah yeah that is what I'm Rajashekar G: au sorry yeah we'll have a clear discussion on what are we 00:39:39   Vara Kumar Jagarapu: thinking okay Rajashekar G: are agreeing upon Vara Kumar Jagarapu: yeah okay yeah I think like you can schedule at 11:30 also that is fine Rajashekar G: we need your inputs here no Vara Kumar Jagarapu: Yeah, we'll think about like as Ashantal suggested right what it is generally people ask right so we can take some examples from the different Rajashekar G: Got it. You got Vara Kumar Jagarapu: projects perspective roles perspective and we can think of one solution or Rajashekar G: it. Vara Kumar Jagarapu: different solutions and we can go for if a is available then we can also ping him yeah that is what I'm thinking Rajashekar G: Let's Vara Kumar Jagarapu: Okay. Rajashekar G: move. Vara Kumar Jagarapu: Yeah. Rajashekar G: Am is also having few Yeshwanth Reddy Yerraguntla: Just before we go there, Rajashekar G: questions. Yeshwanth Reddy Yerraguntla: one more sgestion I can give is you can directly ask uh the notebook LM itself. I think you can use that also as a starting point. just letting me. Rajashekar G: We started there only. Yeshwanth Reddy Yerraguntla: Okay, Rajashekar G: First we started there and we came up with the questions. 00:40:50   Yeshwanth Reddy Yerraguntla: so Rajashekar G: Yeah. So Amul is working on alert life cycle understanding. So majorly it's like yesterday we had a discussion and after that uh so she need to take care of Yeshwanth Reddy Yerraguntla: yeah. Rajashekar G: like how an escalation need to be understandable by an agent. So on that particular point she created a document and she also has some questions. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: I'm on Amulya Maggidi: Yes or no? Rajashekar G: you. Amulya Maggidi: So uh alert life cycle sentiment score. Yeshwanth Reddy Yerraguntla: Yeah, right. Amulya Maggidi: So yeah Yeshwanth Reddy Yerraguntla: H for Gmail, not for everything, but for Gmail, this Amulya Maggidi: for so yes. Yeshwanth Reddy Yerraguntla: is Amulya Maggidi: So sentiment score defining alert life Rajashekar G: and like whoever is working on the Gmail data source agent uh the syncing mechanism Yeshwanth Reddy Yerraguntla: right. Rajashekar G: of the Gmail. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: So there what I have suggested is whenever we are syncing the data into the database. So let's say we are keeping the ID and uh the vectors of that particular 00:42:17   Yeshwanth Reddy Yerraguntla: Makes Rajashekar G: content right so against it we need to have a mechanism in another column where it will Yeshwanth Reddy Yerraguntla: sense. Rajashekar G: be having the sentiment score keywords or the sentiment score. Yeshwanth Reddy Yerraguntla: Correct. 100%. Yeah. Rajashekar G: So there what need to happen is while syncing the data itself whatever the content it's say 100 mails are syncing one after the other one email is synced there need to be another module sentiment module uh needs to be there inside it there will be one LLM analyzing the natural language content and giving me a sentiment Yeshwanth Reddy Yerraguntla: Yeah, you don't need a different LLM. Rajashekar G: score Yeshwanth Reddy Yerraguntla: Uh when you are pulling the data, you're saying uh one LLM analyze. Rajashekar G: yes when syncing the data then only needs to happen Yeshwanth Reddy Yerraguntla: Yeah, this is what I was also thinking. This is this is how I would have done it. Rajashekar G: Okay. So that flow is okay. Yeshwanth Reddy Yerraguntla: Got to cut Rajashekar G: So we are thinking of two ways. 00:43:07   Yeshwanth Reddy Yerraguntla: it. Rajashekar G: One is there will be one column in which it will be having all the keywords sentiment keywords or it will be having a sentiment score.1 to one. Yeshwanth Reddy Yerraguntla: Yeah. No, no, Rajashekar G: We will standardize it like uh how it needs to be done. Yeshwanth Reddy Yerraguntla: no. Do both. Rajashekar G: Do both. Yeshwanth Reddy Yerraguntla: Store. See because we are storing email. Sorry. we are processing email which is already a very costly process. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: We'd rather do as much processing as we can at that one step. So what I suggest is in that one email when you're processing you'll process which people are involved in that email. You'll process which companies or organization or topics are involved in that email. You'll process the tags also. Is this a hiring hashtag hiring hashtag bug hashtag bugfix? These kinds of hashtags also you can generate from the team. Correct. Uh you can generate the sentiment score also. 00:44:04   Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: uh sentiment instead of calling it a score you will say uh what is the sentiment uh uh in one word one like sentiment could be neutral apathy angry joyful celebrating Rajashekar G: All right. Yeshwanth Reddy Yerraguntla: right these are all English words only so if you want one or more English words to represent sentiment of that do that there's no point in storing in float vector uh sorry floating numbers Because floating is always on a scale of 0 to one. And you have to define what that one is. Is one if one is super angry. What if there is another email where some people some person is super sarcastic? Now is that anger or not? You don't. It's just easier to store in text format in hashtag format what is contained in that email. Based on which you can always do quick uh searching. Now you can vector search on the sentiment strings right apathy sentime Rajashekar G: So, Yeshwanth Reddy Yerraguntla: sarcasm these will also sort of get a score on anger scale 00:45:19   Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: automatically Rajashekar G: So how not LM directed us is like there will be a sentiment score less than.3 then you will trigger an escalation. So in that way like what we thought is look now we understood it better like there will be a column in which we'll be having all the tags like which topic it will belong to which project it will belong to everything will Yeshwanth Reddy Yerraguntla: I have multiple columns. That's what I'm saying. Rajashekar G: be and other than one is sentiment column. Yeshwanth Reddy Yerraguntla: Yeah. Correct. Rajashekar G: In sentiment column there will be all the keywords in the particular what are all the emotions we are capturing all those keywords Yeshwanth Reddy Yerraguntla: Correct. Rajashekar G: need to be saved. In another column there will be a score. Let's say we stand something like 0.1 is more angry. Let's say one is more angry, 10 is very happy or appreciation. Based on that we will standardize few things. Based on that we will have a score. 00:46:07   Rajashekar G: So whenever data source agent he is giving the information to RT agent. The RT agent will be knowing like how many escalations there are in these emails. Let's say 100 emails are coming. Yeshwanth Reddy Yerraguntla: Got it. Rajashekar G: How many sentiment columns are less than.3 based on that it will take a Yeshwanth Reddy Yerraguntla: Yeah, Rajashekar G: decision. Yeshwanth Reddy Yerraguntla: makes Rajashekar G: So we need to store everything like multiple it is like a analy analysis module any Yeshwanth Reddy Yerraguntla: sense. Rajashekar G: natural content is coming whether it's coming from Gmail whether it's coming from any other thing if you need to analyze it we will use this module to do this kind of analysis. This LLM is responsible for that whileing the Yeshwanth Reddy Yerraguntla: Got it. Yeah. Rajashekar G: data. Yeshwanth Reddy Yerraguntla: Yeah, it makes sense. Rajashekar G: That one actually not uh represented anywhere Yeshwanth Reddy Yerraguntla: No, Rajashekar G: here. Yeshwanth Reddy Yerraguntla: it's in data agents. When you zoom into the hybrid DB, Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: uh I would have mentioned something like custom indexes/ fields. 00:47:12   Rajashekar G: Yes. Yeshwanth Reddy Yerraguntla: That is where for every data source, we'll define what are those custom fields. Rajashekar G: Okay, Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: sure am are you clear here? Amulya Maggidi: Yes and Rajashekar G: So what I'm thinking is Amula is now working on how a data source Amulya Maggidi: no. Rajashekar G: works and how it how it need to bring the data to RT agent and how that Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: will be synced into our system. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: So entire life cycle not only the core part. Yeshwanth Reddy Yerraguntla: Yeah. Let her do it. Okay. If Arpit and Rahul these people will get pulled into those things anyway it will make Rajashekar G: Yeah. Yeshwanth Reddy Yerraguntla: sense for someone to work on data sourcation. Rajashekar G: Okay. So take your time put your points let's discuss again and Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: in Amulya Maggidi: Okay. Anna Rajashekar G: okay Amulya Maggidi: sentiment. Rajashekar G: right now Gmail Foreign speech. Foreign Yeshwanth Reddy Yerraguntla: See there will there be sentiment for 00:48:16   Rajashekar G: speech. Amulya Maggidi: Okay. Yeshwanth Reddy Yerraguntla: Jira Aalia I'm asking will there be sentiment for Amulya Maggidi: Yes or no? Yeshwanth Reddy Yerraguntla: Jira so what is an so what is an escalation Amulya Maggidi: Uhhuh. No. Based on Yeshwanth Reddy Yerraguntla: rule in Jira now from your perspective Amulya Maggidi: uh it should something uh stay in the Rajashekar G: It's split. Amulya Maggidi: big status uh for too many days or uh something which is already done as again came back to in progress or something like that. Yeshwanth Reddy Yerraguntla: Perfect. Exactly. So every data source will capture different kinds of escalations. Amulya Maggidi: oscillations Yeshwanth Reddy Yerraguntla: Some natural language sources like Gmail will need sentiment. Amulya Maggidi: sentiment Yeshwanth Reddy Yerraguntla: Was that what you were Rajashekar G: Okay. Amulya Maggidi: Oh, yes or no. That's all I Yeshwanth Reddy Yerraguntla: asking? Amulya Maggidi: asked. Rajashekar G: And what I'm expecting is this analysis part like we can say say it as not sentiment analysis we can make it as a natural language analysis that will be a module inside it there will be an LLM sitting customization package a library. 00:49:32   Yeshwanth Reddy Yerraguntla: Yeah. Yeah. Rajashekar G: So whenever in any particular data source syncing is happening we need to use that library to analyze and sync the data in the way we want. Yeshwanth Reddy Yerraguntla: Correct. Amulya Maggidi: and sync the data. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: Go to the data source agent actually. So when you zoom in uh see that blue box right data source or Amulya Maggidi: Oh, Yeshwanth Reddy Yerraguntla: API call someone's data source agent is sending the request data source will give Amulya Maggidi: yes. Yeshwanth Reddy Yerraguntla: back the data so for the last 6 hours these 300 emails were received assume data source agent inside the data source agent there is then one Amulya Maggidi: Yes. Yeshwanth Reddy Yerraguntla: natural language processing module which is processing all the 300 emails and putting custom sentiments score hashtags everything in the hybrid table. Amulya Maggidi: Yes, Yeshwanth Reddy Yerraguntla: So if I have to now update the diagram I come I'm coming Amulya Maggidi: I'm Yeshwanth Reddy Yerraguntla: to the diagram all we are saying is there is an additional uh module Rajashekar G: module. 00:50:44   Yeshwanth Reddy Yerraguntla: now here we are saying data processing module for custom indexes Amulya Maggidi: Okay. So it will have a custom len which only uh use it for uh processing custom indexes. That's it. No. Yeshwanth Reddy Yerraguntla: I mean see that same LLM which is deciding ad hoc question answering can be used for data processing is what I'm saying or maybe it will change right let's not make those comp questions whether it is the same LLM or a different on it. Maybe it might change, but yes, there is a there's a source of intelligence there. Definitely. So, I can even do this if that's what you're asking. Let there be another tiny intelligence there. Rajashekar G: And why I'm saying different means like there will be better uh models right in terms of analyzing the sentiment rather than Yeshwanth Reddy Yerraguntla: M Rajashekar G: going with the same model for everything. Yeshwanth Reddy Yerraguntla: agree. Rajashekar G: Right now we are going with Gemini 2.5 flash for everything like if we are more focusing on sentiment analysis 00:52:00   Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: is there any better LLM for this? Yeshwanth Reddy Yerraguntla: Mhm. Rajashekar G: We need to figure it out and we will use that one here. Yeshwanth Reddy Yerraguntla: Yeah. Makes sense. Makes sense. Rajashekar G: Yeah, either way like but that need to be plug and playable Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: like Yeshwanth Reddy Yerraguntla: Yeah. Yeah. Rajashekar G: configurable. Yeshwanth Reddy Yerraguntla: I mean if if that data processing module you write it properly that one module can be used across different sources. Rajashekar G: Yes. Amulya Maggidi: Okay. Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: Okay. Amulya Maggidi: Yes, sir. Rajashekar G: So next I'm editing mine. So like uh I have combined manufacturing uh this code part code uh separation Yeshwanth Reddy Yerraguntla: Oh, okay. Rajashekar G: part I combined everything into one thing and uh I'm testing it out. Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: Once this is done I'll just make it flag based uh enable disable and give one Yeshwanth Reddy Yerraguntla: Nice. Okay. Rajashekar G: deployment. Then we will merge the pav's Gmail source into this or we'll make 00:53:10   Yeshwanth Reddy Yerraguntla: Okay. Sure. Rajashekar G: a branching strategy like DQ level of this. Yeshwanth Reddy Yerraguntla: V, you should take care of the PR uh review. Rajashekar G: Yes, I'll get it. Yeshwanth Reddy Yerraguntla: Yeah. Vara Kumar Jagarapu: Okay. Rajashekar G: Next Yeshwanth Reddy Yerraguntla: Why are module? Rajashekar G: Rahul is sick. I Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: think Manisha Gundapuneedi: Um, Yeshwanth Reddy Yerraguntla: They don't have any update. It's okay. Manisha Gundapuneedi: huh. Yeshwanth Reddy Yerraguntla: Yeah. Manisha Gundapuneedi: No. Yeshwanth Reddy Yerraguntla: Okay, Manisha Gundapuneedi: I mean Yeshwanth Reddy Yerraguntla: got Rajashekar G: Uh but what I'm saying is data Yeshwanth Reddy Yerraguntla: it. Rajashekar G: source Manisha Gundapuneedi: in Output Rajashekar G: means at least Manisha Gundapuneedi: card Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: And one more thing we are stuck with that experience Manisha Gundapuneedi: experience center. Rajashekar G: part. Manisha Gundapuneedi: So first of dashboard. Yeshwanth Reddy Yerraguntla: Okay. Yeah. Yeah. Manisha Gundapuneedi: So Yeshwanth Reddy Yerraguntla: So, are you using copilot kit? Okay. Manisha Gundapuneedi: yes Rajashekar G: I have some questions there. So if we are using uh copilot kit that copilot kit only for the chart interface Manisha Gundapuneedi: H Rajashekar G: section. Manisha Gundapuneedi: dashboard. So if I go and say it should be able to do and vice versa. with along with some dashboard. Rajashekar G: But before that at least and another question we will It is like what Anita is working on Yeshwanth Reddy Yerraguntla: Heat. Heat. Manisha Gundapuneedi: So Rajashekar G: question. Manisha Gundapuneedi: documentation existing enterprise chat. Replace. Rajashekar G: Okay. Manisha Gundapuneedi: for Rajashekar G: Okay. And anything else? Yeshwanth Reddy Yerraguntla: Level. Vara Kumar Jagarapu: Nothing. Transcription ended after 00:58:32 This editable transcript was computer generated and might contain errors. People can also change the text after it was created.