Mar 16, 2026 Enterprise Brain Scrum - Transcript 00:00:00 Abhilash Adunuri: Second. Rajashekar G: Yes. Abhilash Adunuri: I'm reading Rajashekar G: Clear. Abhilash Adunuri: Okay. 1 2 Rajashekar G: Okay. Amulya Maggidi: morning. Rajashekar G: Okay. Can you make those points Amulya Maggidi: Okay. Rajashekar G: ready? Amulya Maggidi: Uh so Rajashekar G: Hi Yeshwanth Reddy Yerraguntla: Hi. Hi. Rajashekar G: now. Yeshwanth Reddy Yerraguntla: Yeah. Are we waiting for anyone else? Rajashekar G: No. Yeshwanth Reddy Yerraguntla: task. Rajashekar G: Mhm. Yeshwanth Reddy Yerraguntla: Oh my god. Rajashekar G: Actually Yeshwanth Reddy Yerraguntla: Okay. Interesting. Rajashekar G: subtask Yeshwanth Reddy Yerraguntla: Okay. Adi for you. I'm seeing two seven out of seven. For me, Rajashekar G: seven. Yeshwanth Reddy Yerraguntla: I'm seeing Okay. Rajashekar G: Actually this this is mine. So this is Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: a Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: person. Let me head it up. Yeshwanth Reddy Yerraguntla: Yeah. see story-wise uh for you but for me Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: I'm seeing all of them in one Rajashekar G: Yeah. 00:06:53 Rajashekar G: Here we can see here the sprint one Yeshwanth Reddy Yerraguntla: shot. Rajashekar G: we have divided into like two sprints like squared one sprint and square two sprint Yeshwanth Reddy Yerraguntla: Okay. Oh, okay. Rajashekar G: and what pal said what pal said is like per day we need to have seven hours of uh efforts and that will be converted into story points. So right now for this sprint by next week uh Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: 27th uh for the enter team we need to come something something come around 74 story points. Yeshwanth Reddy Yerraguntla: Okay. Okay. Rajashekar G: Okay. Uh based on that we need to modify if there is anything like let's say the first thing is uh let's start one by one person wise person well. So let's say P is there. Ganesh is still working on Yeshwanth Reddy Yerraguntla: Mhm. Rajashekar G: but so what he's working on that is one Manisha Gundapuneedi: Okay. Okay. Rajashekar G: thing and next thing Anita a validate dashboard and chart integration. So what she has done document 00:08:18 Yeshwanth Reddy Yerraguntla: I have reviewed it. Yeah. Rajashekar G: I conclude what we need to give give to her next Manisha Gundapuneedi: Updated document. Rajashekar G: decide. Manisha Gundapuneedi: Sorry. Mannam Sai Anitha: in an item shares up. Yeshwanth Reddy Yerraguntla: by the way um all uh documents put it in chanaka/alpha branch. Yeah. Yeah. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: Let's just maintain a uh gate of like nobody will directly commit to chanaka/ alpha they have to raise PRs but all documents now will show up there as a portal itself is also Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: very fast now um so we'll not have these hangups and anything. Rajashekar G: Okay. Mannam Sai Anitha: Come on a Manisha Gundapuneedi: Open Mannam Sai Anitha: group. Manisha Gundapuneedi: those. Rajashekar G: Next thing Rahul and stars also start and we don't have clarity like what they are exactly Manisha Gundapuneedi: Um Rajashekar G: into Manisha Gundapuneedi: actually documenting Rajashekar G: valid document. Manisha Gundapuneedi: Morning. Rajashekar G: Okay. So I'm moving to Pan has to roll out this Gmail thing to our internal users Friday demo. 00:10:22 Rajashekar G: So these are the points we need to implement and deploy into chanaka.jwami.com Yeshwanth Reddy Yerraguntla: H. Rajashekar G: thewami.com to use the Gmail on enterprise brand for the people like Pratam Naven Sopal Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: we are targeting it in three days like tomorrow has to update the tasks here Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: status next context propagation life cycle Right now enterprise brain is scattered into multiple branches. Yeshwanth Reddy Yerraguntla: Mhm. Rajashekar G: Uh enterprise manufacturing is in one place. Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: Enterprise brain for Salesforce and insurance DB is one place and Gmail we are giving Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: multiple deployments with custom branches. First of all, I want to bring everything into one place and uh make a standardize of branches like dev QA prod. So whatever doing in chaneka also follows that process. Yeshwanth Reddy Yerraguntla: Okay. Okay. Rajashekar G: So I'm working on that. And uh next one uh two stories on your Yeshwanth Reddy Yerraguntla: Both are uh yet to start. Rajashekar G: name. Yeshwanth Reddy Yerraguntla: No actually both are actually in development only. 00:11:48 Yeshwanth Reddy Yerraguntla: First one is in development. Let me move the Rajashekar G: Shall I update? Yeshwanth Reddy Yerraguntla: context. Rajashekar G: Okay. Next one Abilash. So we has given him arg info part in the knowledge base. So he started with LLA document and get approval. He has put the points to you yesterday. Yeshwanth Reddy Yerraguntla: I have given the Rajashekar G: I have also read the Yes. Abby, you have any questions? Yeshwanth Reddy Yerraguntla: feedback. Abhilash Adunuri: only RT agent data source agent like schema how do how do they map to the ID Yeshwanth Reddy Yerraguntla: ID of oh the I Abhilash Adunuri: you mention saying Yeshwanth Reddy Yerraguntla: you're addressing one of the question uh so in both I am not Abhilash Adunuri: yes Yeshwanth Reddy Yerraguntla: clear on how ID works right so uh uh you can discard that point you can just continue with uh uh the schema design like I understood what the ideas mean Abhilash Adunuri: Yeah, what you are saying we need to maintain uh schema in the organ file. Yeshwanth Reddy Yerraguntla: Mhm. So, Postgress has this concept of schemas. 00:13:11 Yeshwanth Reddy Yerraguntla: Or info will have its own schema. Data source agents will have their own schema. So, do not mix both of these in the same uh database. I mean, they are in the same database, but not in the same schema. Abhilash Adunuri: uh not that I mean when I'm reading about world models and concepts what it was saying is uh concepts is uh something related to technical schema let's say if we are using Yeshwanth Reddy Yerraguntla: Correct. Abhilash Adunuri: Salesforce and Jira in the office u all the technical data Yeshwanth Reddy Yerraguntla: Correct. Abhilash Adunuri: schema related to the Salesforce and Jira comes under concepts and the explanation for of that business context will come under world models Yeshwanth Reddy Yerraguntla: Correct. So, Abhilash Adunuri: So, Yeshwanth Reddy Yerraguntla: so uh the concepts for Jira will be separate from concepts for Salesforce will be separate from concepts for org model. Similarly, um runbooks uh skills everything will be separate for all three of them in separate tables. Abhilash Adunuri: Yeah, what our doubt was we will be already having Salesforce schema, GMA schema at data source level, right? 00:14:28 Yeshwanth Reddy Yerraguntla: Yes. Abhilash Adunuri: Should we also make a copy here as well? Yeshwanth Reddy Yerraguntla: No, no, no, no, no. where you have so that is what I'm saying for for data source when you have the world model concepts all these things they will stay there for only the data source those tables are separate what I'm trying to say Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: is cut or Rajashekar G: data source level concepts are different and you're saying uh Yeshwanth Reddy Yerraguntla: level That's Rajashekar G: level source um concepts are different Yeshwanth Reddy Yerraguntla: what I'm Rajashekar G: right but how they are different like uh let's say Yeshwanth Reddy Yerraguntla: saying. Rajashekar G: there is a data source called Jira or Gmail and sales the concepts Yeshwanth Reddy Yerraguntla: Okay. Okay. Rajashekar G: means like uh uh it's schema Yeshwanth Reddy Yerraguntla: H Jira schema or Salesforce schema. Rajashekar G: Right. What's here Yeshwanth Reddy Yerraguntla: Here the concepts will be specific to real time model. Rajashekar G: then? Yeshwanth Reddy Yerraguntla: Right? What do we mean by diwami as an organization? 00:15:33 Yeshwanth Reddy Yerraguntla: Um what are projects like? What are pipelines in the organization? It's you can say okay why are these concepts? These are not concepts. These are world model. Sure. In that case yeah concepts can be empty. Rajashekar G: Okay. Abhilash Adunuri: Yeah, I'm honest. Rajashekar G: And the real picture is real picture is Abhilash Adunuri: We thought Rajashekar G: uh how many projects are there in Diwami? How many people are there in Diwami? What are the roles? Aan. Yeshwanth Reddy Yerraguntla: H correct. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: So if I take an example uh I will just take one or two minutes. Uh let me share my screen. So for us the whole reason of having concept world models all those things is in our chanaka service we might have something like uh chanaka uh create new agent command line um uh command line uh argument function where uh you'll say source sorry uh we'll say something like uh unifi If I do ML point is this YML might contain JSON this JSON might contain what is the world model right what is the what are the concepts they we are dealing with what are the skills run books blah blah blah 00:17:18 Rajashekar G: Skills means tools here. Yeshwanth Reddy Yerraguntla: H correct. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: So when that happens when this when this runs immediately what should happen is I should create a fresh uh schema in posgress pgsql. There is a new schema that is going to be created as unifier_ um data source agent uh schema. In that schema you're automatically creating uh world model concepts skills the maybe these are all separate tables or separate uh whatever columns I don't know uh runbooks could be um I still am not clear on runbooks for one thing because this should be code right it's not just text file so we shouldn't store code in posgress in any way so how will that work like if our run hooks going to be uh referenced via folder names right here dot run sorry skills run books dot skills some kind of folder location you're giving saying this in this folder all the runbooks sorry skills are Right. So the moment you the run this one guy lot of uh automatic scaffolding happens. 00:18:53 Yeshwanth Reddy Yerraguntla: So that um next time when you simply launch uh Chanaka uh run data source agent for ID. Sorry. Yeah. ID equal to unifier name equal to unifier or ID whatever you want to call it. The moment you do this, it will launch uh you know in 8,000 in this lower location. It is directly launching the pylog UI uh local host 18093. It is launching A2A server and you can start testing it and the moment you start chatting that agent already knows I have to go here and fetch my system prompt. when user is asking certain questions I have to come here and identify the skills all other things are also there right you have rag question qa you have training something uh skills common training skills am I making sense Rajashekar G: Yes, Emma. Yeshwanth Reddy Yerraguntla: So the fact that from this one guy we are able to create separate tables that is what we mean by clean separation of all data source agents having their own world models all those things. 00:20:28 Yeshwanth Reddy Yerraguntla: So the same is applicable for the uh or argent also. So here we might say type equal to data source or type equal to real time which is the one and only kind of a thing. Rajashekar G: Uh but this is like we are initiating a data source agent Yeshwanth Reddy Yerraguntla: This is what this is for a data Rajashekar G: right? Yeshwanth Reddy Yerraguntla: source. Correct. Rajashekar G: Okay. All these things going and sitting in knowledge base where RT agent is having the communication. Yeshwanth Reddy Yerraguntla: No, none of these will go and sit in knowledge base. That's what I'm saying. Rajashekar G: Okay. So all these are data source agent Yeshwanth Reddy Yerraguntla: Correct. for knowledge base specifically. Rajashekar G: only. Yeshwanth Reddy Yerraguntla: Um here where are we for knowledge base? See you have to come up with u the right calls here because you already know how diwami works. If you think only world model is sufficient and concepts is not needed, you can keep it blank. 00:21:45 Yeshwanth Reddy Yerraguntla: The point is because I created a realtime agent, all these tables are automatically getting created except the user information. This we'll take care later. This this is also something that should uh talk to realtime agent, right? So there is a real-time agent specific skill to read and update snapshot. Point is the moment you define those end points which 340 these kinds of endpoints whatever you are defining they will be applicable for the other guys also. Am I am I clear or am I confusing you guys? Abhilash Adunuri: Yeah. Can I say uh concepts nothing but the uh business terminology Yeshwanth Reddy Yerraguntla: Repeat what you Abhilash Adunuri: Dwami use? Yeshwanth Reddy Yerraguntla: Okay. Abhilash Adunuri: Can I can I conclude in that way so Yeshwanth Reddy Yerraguntla: Continue. Abhilash Adunuri: that Yeshwanth Reddy Yerraguntla: World model is the business aspect of it. Abhilash Adunuri: Yeah. Yeshwanth Reddy Yerraguntla: Concepts will be the technical aspect. So I have a feeling that at least for the real time agent maybe we don't need any concepts. 00:23:07 Yeshwanth Reddy Yerraguntla: World model should be sufficient and world model like I said is nothing different from a system prompt. See anyway you start with creating these tables even if concepts is empty it's okay for the time being. Yeah, just like get started on the hands-on activity, create them and get a feel of how if you were to talk to realtime agent by yourself directly. What kind of questions will you ask? You'll say give me the latest news from yesterday or what happened in Salesforce yesterday or what meetings happened yesterday. Those kinds of questions if you ask how to get answers what are the kinds of things you have to feed in your world model and concepts ask yourself backward Rajashekar G: Uh if that is the case continue you have questions. Yeshwanth Reddy Yerraguntla: Chip. Rajashekar G: So if like any ADO query comes from the user RT Yeshwanth Reddy Yerraguntla: Yeah, Rajashekar G: agent first it will check in its knowledge base whether it has the answer or not. Yeshwanth Reddy Yerraguntla: you're right. Rajashekar G: then ask the data source agent if it is not find 00:24:31 Yeshwanth Reddy Yerraguntla: Correct. Rajashekar G: anything. Okay. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: And yesterday while we were discussing we were not clear like what to store in the AG snapshot exactly. Yeshwanth Reddy Yerraguntla: Um was that addressed here like in this document snapshot Rajashekar G: No. Uh that one is different. Yeshwanth Reddy Yerraguntla: algorithm? Okay. Abhilash Adunuri: has to be different. Yeshwanth Reddy Yerraguntla: uh think of our project only, right? What do we mean by snapshot? Since yesterday, what happened? What are all the sources of information uh that can help me create a snapshot? In my mind, there are two different ways of creating the snapshot. One is a simple text file saying these are all the things that happened yesterday. The other way is to actually use a graph data structure where you have nodes and uh stuff like that. Because we are still in the beginning stages, I will say let's probably deal with the text file only. So currently the snapshot could be one text file for every project that happens and um in the text file everything about the latest information should be there. 00:25:49 Yeshwanth Reddy Yerraguntla: However big that text file Rajashekar G: So the text file means uh the file will be stored in server and uh are we pointing to that one Yeshwanth Reddy Yerraguntla: is Rajashekar G: in the DB or in DB itself we will have a column of uh uh raw text. Yeshwanth Reddy Yerraguntla: we'll have a column of raw text for that date and Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: uh um yeah that will be a good starting point. Now if that itself is very inefficient in the sense that every time I have to read a full Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: text file can't I do vector search those are the kinds of answers we can get once we have a v1 ready at least we know how um how to build something after that we can worry about the uh the complexities and efficiency and these kinds of My point is designing the snapshot in itself is a very Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: um unknown very big unknown primarily because no one in the world is doing this and even if they're doing it's very it's not in open source in any way. 00:27:07 Yeshwanth Reddy Yerraguntla: So we have to invent a lot of uh things here. So one guideline I will give is here I kept on I kept asking a couple of things Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: like give examples give examples. Okay. So I think I saw one or two examples at the very bottom but try to have those examples also like at the top when you're saying what is the purpose of this thing right? you you did use the word purpose properly that this is why you want to store whatever it is or below that purpose you also had enough examples of how it is applicable in uh enterprise brain for that concept. So our world models and concepts applicable for data sources for realtime agent how are runbooks applicable for realtime agents and data sources. Same thing will happen in other LLDs also that if you're dealing with ARG snap snapshot what is the purpose how will it look like what kind of questions can I answer with the snapshot they will help both the reader and the writer understand what they are trying to deal with in in somewhat technical terms these are uh use cases right so these are functional requirements um functional requirement sort of means you get into the business territory of why you are building this. 00:28:41 Yeshwanth Reddy Yerraguntla: So you can also try to write uh uh dedicate at the very top what are the functional and non-functional requirements for that uh concept. Is that clear? not clear. Rajashekar G: It's clear enough for me. Yeshwanth Reddy Yerraguntla: All I'm saying is because you're dealing with your this nobody wrote this manually, Rajashekar G: Abby Yeshwanth Reddy Yerraguntla: right? You made LLM do it. So ask the LLM also to create functional and non-functional requirements at the very top. They will give us proper idea of whether the document is in the right track or not. Also that is missing. And second thing that is missing is how do we test the whole uh pipelines right whatever you build how do we test them that is also needed. So once these two are also ready these first section and last section directly Prabhakaras can consume and tell is this on the right track or not because these are more business oriented right need a business off Rajashekar G: That's Yeshwanth Reddy Yerraguntla: also. 00:29:57 Abhilash Adunuri: Yes, I'm a Rajashekar G: right. Yeshwanth Reddy Yerraguntla: Yeah. So I have these two also updated. So I have seen um just comments sorry I have seen this one thing alert life cycle uh whoever created uh uh you created in the wrong location you uh I think uh you created Rajashekar G: Hi s***. Yeshwanth Reddy Yerraguntla: here it should have been inside uh LLDS because when I click on this guy Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: it's clearly telling me it's DB353 3 right and there is a file for deb 353 35 just replace it or Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: yeah do not take this as the ground truth in any way this is just a placeholder that I had the Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: other document is copilot kit guide I have seen this um I have only one question for uh uh Anita if I open up uh two ways to deploy mode to self-hosted runtime right so are you saying we don't need to rely on copilot in any Okay. Mannam Sai Anitha: URLs 00:31:28 Yeshwanth Reddy Yerraguntla: Okay. Mannam Sai Anitha: component interface Yeshwanth Reddy Yerraguntla: Okay. Mannam Sai Anitha: and then Assistant message, user Yeshwanth Reddy Yerraguntla: Got it. Right. Mannam Sai Anitha: message. Yeshwanth Reddy Yerraguntla: So, so see this is in our hands. This is in our hands because we are writing it. Mannam Sai Anitha: Awesome. Yeshwanth Reddy Yerraguntla: Your back end fast APR express is also in our hands because we are writing it. Mannam Sai Anitha: Oh, I Yeshwanth Reddy Yerraguntla: If this is also in our hands that all you have to do is get clone copilot kit and run npm Mannam Sai Anitha: see. Yeshwanth Reddy Yerraguntla: rundev with copilot kit components. Uh like that sentence is that correct that we can do npm rundev just by git cloning the whole thing and not rely on their servers in any way. Mannam Sai Anitha: Foreign speech. Foreign speech. Yeshwanth Reddy Yerraguntla: Okay. Uh now I leave it to Manisha and Ganesh to take the call. Do you want to build uh everything from scratch or do you want to take advantage of copilot kit? 00:32:47 Yeshwanth Reddy Yerraguntla: Because if this is something that we don't need to rely on them. Uh it makes sense for me to uh you know take advantage of it. Rajashekar G: But pricing will be there now. Yeshwanth Reddy Yerraguntla: Why will it be there when you're doing get clone of copilot kit and running it locally? Why will it be there? Rajashekar G: I think there is an agent sitting inside, right? We have seen like uh an API key we were using. Yeshwanth Reddy Yerraguntla: No, but that Mannam Sai Anitha: back in place of public Rajashekar G: Okay. Mannam Sai Anitha: key. Yeshwanth Reddy Yerraguntla: should Rajashekar G: Okay, got it. Mannam Sai Anitha: Oh. Rajashekar G: So all the UI things uh this kit will take care of. Mannam Sai Anitha: Oh. Yes or no? Rajashekar G: So will it include all the visualization points also like visualization related widgets and Mannam Sai Anitha: No. Rajashekar G: everything? Mannam Sai Anitha: No. Only chart related Yeshwanth Reddy Yerraguntla: visualization related things we have to write it. Rajashekar G: Okay. Mannam Sai Anitha: components. Yeshwanth Reddy Yerraguntla: No, we once we write it will decide 00:33:59 Rajashekar G: Yeah, Ganesha is already working on that experiential part, right? So, Yeshwanth Reddy Yerraguntla: That's what if Ganesh is confident that this is all unnecessary again we'll waste time reading and Rajashekar G: yeah, Yeshwanth Reddy Yerraguntla: understanding a separate library which may or may not break in the future. Rajashekar G: got Yeshwanth Reddy Yerraguntla: I'm happy to let Ganesh take the call and uh work on the UI because from the document Rajashekar G: it. Yeshwanth Reddy Yerraguntla: that I read it's seems to be relatively straightforward how copilot kit works. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: Yeah, Manisha Gundapuneedi: Okay. No, sorry. Today we'll just sit and we'll take a Yeshwanth Reddy Yerraguntla: you can uh take a decision with uh Gopala or Pra not Manisha Gundapuneedi: decision. Yeshwanth Reddy Yerraguntla: Gopal sir. Yeah. Show this document and use it. Um if that is all then we can drop off if there are no other blockers. Amulya Maggidi: And now I have some questions. Yeshwanth Reddy Yerraguntla: Okay. Do you want to share screen or something Mannam Sai Anitha: still. 00:35:12 Amulya Maggidi: Oh no. Yeshwanth Reddy Yerraguntla: or Amulya Maggidi: No. This one. generally alert. So escalation rules but already data source Yeshwanth Reddy Yerraguntla: Uh, true. What's the question? Amulya Maggidi: already data sources alerts es Yeshwanth Reddy Yerraguntla: So, Amulya Maggidi: data source agent to Rajashekar G: So, And I'll just repeat once. Yeshwanth Reddy Yerraguntla: okay. Rajashekar G: So what is the point is like uh what we have divided is one is uh core agent part and another one is data source agent part. Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: Right? In core agent life cycle we are having a uh thing like alert life cycle back end and Yeshwanth Reddy Yerraguntla: Correct. Rajashekar G: alert life cycle agent to understand escalation these two parts. So in our diagram like uh the data source agent Yeshwanth Reddy Yerraguntla: Right. Rajashekar G: itself telling me RT agent what is information what is alert what is insight and what is escalation. Yeshwanth Reddy Yerraguntla: Correct. Rajashekar G: So why in core thing again we need to come up with these things and plus that need to be taken care by data 00:36:46 Yeshwanth Reddy Yerraguntla: Okay, fair Rajashekar G: source agent while writing the data source agent itself we need to taken care that which columns will Yeshwanth Reddy Yerraguntla: enough. Rajashekar G: decide which is an alert which is an escalation and everything right so our clarity goes Amulya Maggidi: So I Yeshwanth Reddy Yerraguntla: So, continue. Continue. Okay, that's a valid point that if these guys are already dealing with alerts, realtime agent in our current understanding, it will simply route the agent to the user interface. That's what you're saying. Rajashekar G: Yes, Yeshwanth Reddy Yerraguntla: But what Rajashekar G: the real team agent will read this JSON if alerts or escalations is Yeshwanth Reddy Yerraguntla: is Rajashekar G: there then only it will trigger some alert like mail mail Yeshwanth Reddy Yerraguntla: Yes. Rajashekar G: notification or whatever it will just trigger it is only communicating whatever comes summarizing and updating it snapshot that's it so it is not deciding what is an alert Yeshwanth Reddy Yerraguntla: It is not deciding correct. But what it is actually doing is because it will gather multiple Rajashekar G: Yeah. 00:37:51 Yeshwanth Reddy Yerraguntla: alerts from different sources. It might uh it might has it might have to like Rajashekar G: Mhm. Yeshwanth Reddy Yerraguntla: synthesize different different u alerts from different sources into one big uh uh message and send it to user aware agent. Secondly, user aware agent is deciding for every user in the whole system how to send the messages. Correct? Rajashekar G: Uh we have kept this point here. Now one more thing art agent itself can trigger alerts right? Yeshwanth Reddy Yerraguntla: where Rajashekar G: Why it need to again give it to user aware agent and this Yeshwanth Reddy Yerraguntla: because if the alert is like someone in in uh so and so some so and so client is angry simply alert came from Gmail. Now that that particular client is angry not everybody in Diwami needs to know. So how will realtime agent know whom to send the alert to? Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: So that is why it is only figuring out okay I received five alerts from five different sources. Let me see the snapshot. 00:39:09 Yeshwanth Reddy Yerraguntla: Let me understand the context. Let me come up with the alert story. See alert is not just that someone is angry, right? Rajashekar G: H. Yeshwanth Reddy Yerraguntla: The alert should be all about like what was happening since 2 days? Why did something escalate? like it should build a story by itself so that you are once you read the Gmail or Rajashekar G: You're right. Yeshwanth Reddy Yerraguntla: WhatsApp you're fully on board with why that escalation happened in other words if I simply see in my WhatsApp customer Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: is angry again I have to come to Chanaka and ask it which customer is angry why is he angry what happened so if it can come up with the full story and give it to user aware agent user aware agent will do the next step which is okay. I got the full story from the story. Sometimes that story may have finances. I should not send it to uh uh team leads. Sometimes the story might have very low-level information like bugs. 00:40:07 Yeshwanth Reddy Yerraguntla: I don't need to send that to Pratima ma'am because she is into customer satisfaction. That those decisions user aware agent is taking care. So that is also still part of the life cycle. Rajashekar G: eight. Yeshwanth Reddy Yerraguntla: So if if the question is because alerts are anyway coming from data source agents what do I need to do? What I'll say is take ownership of the full life cycle. Let alerts be created from data source agents. You decide how the uh how the payload is going to be received by RT agent. Even before that you decide how to store escalation criteria in data source agents. This is for Rahulia. You decide escalation rules. How to store the escalation rules. How to trigger escalation uh and in what format to the realtime agent? How will realtime agent consume those uh escalations? How will it synthesize news? Should it use a skill? Should it use a runbook to synthesize that story? 00:41:15 Yeshwanth Reddy Yerraguntla: Then when it is sending to user aware agent, how will it uh in what packet or what payload it will send? What JSON key will it use? Then how will you how will user aware agent consume it and send it to the user interface in the form of Gmail or WhatsApp. So many decisions you have to still take. So just taking touching all these points will itself give you enough uh work for the sprint. Amulya Maggidi: Yes or no? Got Yeshwanth Reddy Yerraguntla: So your work is now diffusing from just the core to data source agents also which I think is Amulya Maggidi: it. Okay. Yeshwanth Reddy Yerraguntla: okay Amulya Maggidi: So I need to define uh escalation alert for every schema like every data source agent. Yeshwanth Reddy Yerraguntla: you will define huh once you define the schema saying this is how I want to store escalation rules when you create the data source agent with that you saw this right here we missed one thing now here we are we didn't uh you can talk about escalation rules. 00:42:17 Yeshwanth Reddy Yerraguntla: So you can add that this is how a user has to create escalation rules. This is this a text file is this a JSON? I don't know where it will go and sit in posgress database and Amulya Maggidi: Okay, got it. Yeah. Yeshwanth Reddy Yerraguntla: in what format next when you um run data source Amulya Maggidi: Okay. Yeshwanth Reddy Yerraguntla: agent and when you ask a question question that is very like you know towards the in the direction of give me all escalations will it properly give escalations and in what payload should it give Amulya Maggidi: Okay. What format should it return? That should also be defined. Yeshwanth Reddy Yerraguntla: You should define that and once that format is frozen realtime agent should be in a position to consume that Amulya Maggidi: Yeah. Yes. Yeshwanth Reddy Yerraguntla: format. Amulya Maggidi: Then it should uh sync with and correlate with the organization snapshot and uh correlate to what users and uh combine all the alerts. That's it. Now, Yeshwanth Reddy Yerraguntla: Perfect. Amulya Maggidi: yeah. 00:43:17 Yeshwanth Reddy Yerraguntla: Yes. Amulya Maggidi: then it should send it to user aware agent. Then user aware agent has to make some rules or something so that it can uh pass the alert to whatever user that they can consume. Yeshwanth Reddy Yerraguntla: So the same alert can go to multiple users in different formats. Amulya Maggidi: Yeah. Yeah. Yeshwanth Reddy Yerraguntla: That is that is the key uh distribution Amulya Maggidi: This of user. Yeshwanth Reddy Yerraguntla: activity. Amulya Maggidi: Okay. Yeshwanth Reddy Yerraguntla: So at that point you have to figure out how to uh do arbback. Maybe arbback is too much of effort now. So you can keep it as out of scope. Let me just send information to all users. So you define the testing criteria. I'm going to write this code. Amulya Maggidi: Okay. Yeshwanth Reddy Yerraguntla: I'm going to test it this way. If I talk with data agent, this is how it is going to respond. If I talk with realtime agent, this is how it's going to respond. 00:44:11 Amulya Maggidi: Okay. Yeshwanth Reddy Yerraguntla: If I talk with user aware or agent, this is how it will respond. that kind of uh promises you give in the document saying this is how I'm going to test it then uh prair I will um give further uh Amulya Maggidi: Okay. Yes, sir. Yeshwanth Reddy Yerraguntla: clarifications. Rajashekar G: Okay. Now, Amulya Maggidi: Oh god. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: okay. Kalakonda Harshith Rao: Hi now uh actually I did for the alert life cycle back end once please check and verify is there any I missing ideology behind what I wrote Yeshwanth Reddy Yerraguntla: LLD for this. Rajashekar G: actually hardship part is different now alert life cycle back end so which is Kalakonda Harshith Rao: uh Yeshwanth Reddy Yerraguntla: Oh, okay. Oh, Rajashekar G: his only part is he will build a module where uh there will be some APIs that can able to Yeshwanth Reddy Yerraguntla: yeah. Rajashekar G: trigger emails and WhatsApp notific Okay, Yeshwanth Reddy Yerraguntla: Okay. thing but anyway I'll go through in 2 minutes. Kalakonda Harshith Rao: Sounds Yeshwanth Reddy Yerraguntla: So alert notification alert triggered agent or back 00:45:14 Rajashekar G: that should Yeshwanth Reddy Yerraguntla: end event. Okay, you have to write down then which agent that let that be also clear and what Rajashekar G: All Yeshwanth Reddy Yerraguntla: do you mean back end event? Which back end event in the whole diagram, right? Rajashekar G: right. Yeshwanth Reddy Yerraguntla: Identify. Kalakonda Harshith Rao: heat. Yeshwanth Reddy Yerraguntla: Okay. Identify logged in user from O token. Got it. Create alert. This probably is wrong because as we discussed if okay so from what I understood with Amoleia's discussion the moment Amoleia sends this this blue line her job is over. Rajashekar G: Your screen is not sharing. Yeshwanth Reddy Yerraguntla: Oh sorry. Oh I was not sharing since then. Rajashekar G: No, Yeshwanth Reddy Yerraguntla: No. Rajashekar G: my screen was sharing. Yeshwanth Reddy Yerraguntla: Okay. Okay. Uh sure. Rajashekar G: Stop. Yeshwanth Reddy Yerraguntla: Yeah. Not bad. Um the moment Anita sends something sorry Anita sends something her job is over and he from here onwards the back end takes care of sending the in information to user. 00:46:22 Yeshwanth Reddy Yerraguntla: Correct. Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: That's what Harsh is looking at. Kalakonda Harshith Rao: Yes or no? Yeshwanth Reddy Yerraguntla: Basically this at this point her scope is over and hashid scope will Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: start which means both of you have to Kalakonda Harshith Rao: Yes. Yeshwanth Reddy Yerraguntla: decide uh and on the correct payload format first of all because Anita sorry because Amalia is saying Rajashekar G: I'm working. Yeshwanth Reddy Yerraguntla: uh uh this is the here is an agent that will decide for every user what to And you don't even need to worry about uh authentication in my head. You are already getting for every user what message needs to be received. Kalakonda Harshith Rao: Okay. Yeshwanth Reddy Yerraguntla: So figure out what is that payload with her. Yeah, sure. Create an alert in DB uh based on the message uh packet. You will figure out how to send email, WhatsApp or browser notification. Done. This this the bottom half is correct. Top half you have to be very clear. 00:47:22 Yeshwanth Reddy Yerraguntla: Alert states is also fine. Kalakonda Harshith Rao: Okay. Yeshwanth Reddy Yerraguntla: DB schema, I don't know. I don't know why we need this. If it's an alert, it's an alert, right? Kalakonda Harshith Rao: But antenna know the status of the alert like which severity it's coming from from the Yeshwanth Reddy Yerraguntla: No. Kalakonda Harshith Rao: payload. Yeshwanth Reddy Yerraguntla: So that's the thing like do do we want the agent to decide whether something is high, medium, critical? See, maybe we can say it is critical or non-critical. But even for the scope of the current document everything is critical. Kalakonda Harshith Rao: Okay. Yeshwanth Reddy Yerraguntla: Then tomorrow we might say okay let me also get non-critical alerts. Then the user will start deciding what is a non-critical alert. That is phase two. Kalakonda Harshith Rao: Okay. Yeshwanth Reddy Yerraguntla: For me non-critical could be people have uh created a fresh document in viasa. For praaser it's not right. So this criticality is coming from a user preference but 00:48:20 Kalakonda Harshith Rao: Yeah. Yeshwanth Reddy Yerraguntla: actual critical messages is something that is coming from the data source agent. You decided already that something is super critical. So probably this is the severity column is not needed. Then you have source then you have status uh create okay Kalakonda Harshith Rao: Okay. Yeshwanth Reddy Yerraguntla: result again once you send a message who is calling up and seeing its result Kalakonda Harshith Rao: Like uh actually we are uh planning for the browser notifications where Yeshwanth Reddy Yerraguntla: right Kalakonda Harshith Rao: we give an enable or sorry we can see read all notification where we can get a flag where we if the user haven't seen any alert the mark will be there if if he sees that okay it's opened and it's been seen by the user so we can change the status in DB so Yeshwanth Reddy Yerraguntla: Okay, Kalakonda Harshith Rao: we just show them Yeshwanth Reddy Yerraguntla: maybe you can push it to phase two for the time being. Let let you uh let the system send ah sure okay Rajashekar G: That's 00:49:22 Kalakonda Harshith Rao: Yeah. Rajashekar G: onion. Yeshwanth Reddy Yerraguntla: then uh then probably write it down channels meta data created at updated APIs sorry I don't think I need to give lowle these things here retry on failure is also fine security subtasks also I think you can take care again so a Rajashekar G: Sure. Kalakonda Harshith Rao: Sure. Yeshwanth Reddy Yerraguntla: few things that are missing are functional requirements, non-functional requirements and testing uh criteria Kalakonda Harshith Rao: Testing criteria is last end Yeshwanth Reddy Yerraguntla: testing integration tests. Okay. Kalakonda Harshith Rao: testing. Yeshwanth Reddy Yerraguntla: Unit tests mock and gr SMTP test and Rajashekar G: Heat. Yeshwanth Reddy Yerraguntla: mock studio has no phone correct but better to take prair opinion end to end test Kalakonda Harshith Rao: Yes or no? Yes or no? Yeshwanth Reddy Yerraguntla: post alert/ send with end to end is not this right end to end will be what amia will be sending this is not end to So you have to update Kalakonda Harshith Rao: Huh? Okay. But but know when we are want to send WhatsApp we we need to get API from whether either 00:50:35 Yeshwanth Reddy Yerraguntla: the Kalakonda Harshith Rao: from WhatsApp business or Trellio. So, Trilio is a is a uh purchasable AP. Yeshwanth Reddy Yerraguntla: Okay. Kalakonda Harshith Rao: It will get cost. Rajashekar G: see that we will take care like already this is a solved problem where infinitism is currently Kalakonda Harshith Rao: Okay. Rajashekar G: using that we will discuss and decide what to use what not to use why because already Kalakonda Harshith Rao: Sure. Rajashekar G: in some projects we are using WhatsApp also already and email notifications also people are sending Kalakonda Harshith Rao: Okay. Yeshwanth Reddy Yerraguntla: Nice. Rajashekar G: in other Kalakonda Harshith Rao: people email is good and as I know as I didn't check no Rajashekar G: projects Kalakonda Harshith Rao: problem thing sure that's for Yes, sir. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: Yeah, great. Rajashekar G: Clear. Amulya Maggidi: For now I'm Rajashekar G: Okay. Sure. So, right now I can see there is a gap in like uh data sourced stuff. We haven't started anything like uh rack cache. Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: Okay. carpet is looking at exactly 00:51:52 Yeshwanth Reddy Yerraguntla: Yeah. But this will help you anyway, Rajashekar G: like Yeshwanth Reddy Yerraguntla: right? To come up with this paradigm, Rajashekar G: Yes. Yeshwanth Reddy Yerraguntla: you'll anyway you can anyway create some dummy data source agents and start working. Rajashekar G: Yeah. Can you share this in the chat? This text Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: file. Yeshwanth Reddy Yerraguntla: Second thing is I think uh to some extent I wrote um what is this Jira agent right? Jira skill is different by the way from Jira agent. Uh where is Chanaka back end? Chanaka Jira right to some extent I wrote some mock uh these things. So can always use these as starting points and build on them. Rajashekar G: Okay. Amulya Maggidi: Last one more thing. Uh so escalation rules. So data source rules different. Yeshwanth Reddy Yerraguntla: Yes, they will be different. Amulya Maggidi: So Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: Correct. Dani, you have to define world model. You have to define concepts that that manual work you have to do. 00:53:19 Rajashekar G: Abash Amulya Maggidi: Okay. World models in the concepts. Okay. Rajashekar G: will take care rules work like how a data source need Amulya Maggidi: Only S. Rajashekar G: that escalation you need to do the RN and put those points but Amulya Maggidi: Okay. Yes. Rajashekar G: ech system like while while saving the world model concepts itself we need to save the escalation rules also Amulya Maggidi: Yes, Rajashekar G: somewhere for that particular data source We need Amulya Maggidi: son. Yes, sir. Got Rajashekar G: to figure it out. Amulya Maggidi: it. Rajashekar G: Like you can also discuss with Abilash how he is looking at things and Amulya Maggidi: Okay, good. Rajashekar G: sure you can ask. No Amulya Maggidi: for no clearing. Rajashekar G: problem. Yeshwanth Reddy Yerraguntla: Okay, I can also set up another half an hour meeting in the evening every day. Amulya Maggidi: Oh, Yeshwanth Reddy Yerraguntla: Yeah, Rajashekar G: H. Amulya Maggidi: okay. Yeshwanth Reddy Yerraguntla: be like end of the day anyway. So it's not like you'll do anything additional. Certainly I think at the beginning of the day is sufficient. 00:54:33 Rajashekar G: Sure. If anytime needed we will put in the group itself like we Yeshwanth Reddy Yerraguntla: Put in the group directly. Rajashekar G: need. Yeshwanth Reddy Yerraguntla: Call me on WhatsApp also if you think uh you want. Yeah. Rajashekar G: Sure. Sure. Manisha Gundapuneedi: Anamo Yeshwanth Reddy Yerraguntla: H everyone else can drop off. Uh I'll just sit with her and see how to fix it. Kalakonda Harshith Rao: Zeno. Yeshwanth Reddy Yerraguntla: Thank you. Bye. Have fun. Manisha Gundapuneedi: Sorry. Yes. Yeshwanth Reddy Yerraguntla: slash admin / Manisha Gundapuneedi: Hello Yeshwanth Reddy Yerraguntla: what is this? We I'm saying for some reason um opt. So if I impersonate Manisha at the redwami.com. What is your email address? Manisha Gundapuneedi: N E Yeshwanth Reddy Yerraguntla: Rend address money. Manisha Gundapuneedi: Open Yeshwanth Reddy Yerraguntla: Save. Save in the Manisha Gundapuneedi: it. Yeshwanth Reddy Yerraguntla: JSON character. So let me refresh. Okay, I have to impersonate a different man login. 00:56:47 Manisha Gundapuneedi: Just not just Yeshwanth Reddy Yerraguntla: What admin? Sorry. Manisha Gundapuneedi: Are you Yeshwanth Reddy Yerraguntla: I mean, yeah, Manisha Gundapuneedi: ready? Yeshwanth Reddy Yerraguntla: it it will show up that you have access. No login check. Oh s***. Manisha Gundapuneedi: Heck, Yeshwanth Reddy Yerraguntla: Something is off. Oh, that's because I impersonated you. It got uh understood. Manisha Gundapuneedi: snacks are in. Yeshwanth Reddy Yerraguntla: Clear. I know what it is. Stop impersonation. save our back. Yeah. Mini say start log out login and it should work. Manisha Gundapuneedi: The side does not support now. Yeshwanth Reddy Yerraguntla: Yeah. Manisha Gundapuneedi: log Yeshwanth Reddy Yerraguntla: Can there bottom left? Manisha Gundapuneedi: production Yeshwanth Reddy Yerraguntla: What kind? All right. Arbback law. So everyone has Manisha Gundapuneedi: Okay. Yeshwanth Reddy Yerraguntla: default Manisha Gundapuneedi: First JSON are back Yeshwanth Reddy Yerraguntla: first JSON loan. Manisha Gundapuneedi: related. Yeshwanth Reddy Yerraguntla: I was trying to change it with a different user impersonation just to see how other people will see. Manisha Gundapuneedi: Okay. Yeshwanth Reddy Yerraguntla: So when I impersonate you and then try to change our back, it will stop me. Manisha Gundapuneedi: Okay. Okay. Thank you. Transcription ended after 01:00:02 This editable transcript was computer generated and might contain errors. People can also change the text after it was created.