Mar 2, 2026 Enterprise Brain - V2 - Arch - Transcript 00:00:00 Yeshwanth Reddy Yerraguntla: Hello. Manisha Gundapuneedi: I know. Yeshwanth Reddy Yerraguntla: I don't know. I want to show Satyasri Prabhakar Mantripragada: connect with he started Yeshwanth Reddy Yerraguntla: Yeah. Satyasri Prabhakar Mantripragada: working so was on leave last week that he'll have to support us during Yeshwanth Reddy Yerraguntla: Okay. Satyasri Prabhakar Mantripragada: your night diagram then he should be able to figure it out but overall we can guide him on the engineering standpoint. point that Yeshwanth Reddy Yerraguntla: Yeah. Satyasri Prabhakar Mantripragada: help. Yeshwanth Reddy Yerraguntla: Sorry. Manufacturing. Satyasri Prabhakar Mantripragada: Okay. Okay. Yeshwanth Reddy Yerraguntla: It's 6:15. Why not? So instead of developing What do you call me? Manisha Gundapuneedi: Okay, sir. Satyasri Prabhakar Mantripragada: Okay. And by the way, Yeshwanth Reddy Yerraguntla: Uh-huh. H. Satyasri Prabhakar Mantripragada: contract contract to be reviewed SAP system. Yeshwanth Reddy Yerraguntla: Uh, spelling Okay. Manisha Gundapuneedi: Foreign speech. Foreign speech. Yeshwanth Reddy Yerraguntla: Okay. Satyasri Prabhakar Mantripragada: proposal. Take something. Manisha Gundapuneedi: How do Yeshwanth Reddy Yerraguntla: Okay. 00:06:51 Yeshwanth Reddy Yerraguntla: So Manisha Gundapuneedi: Oracle unifi? Yeshwanth Reddy Yerraguntla: yeah. Manisha Gundapuneedi: forcly Yeshwanth Reddy Yerraguntla: All right. Manisha Gundapuneedi: platform. Yeshwanth Reddy Yerraguntla: Oh no. Manisha Gundapuneedi: Enterprise brain and visual effect. Yeshwanth Reddy Yerraguntla: H Manisha Gundapuneedi: Um Google Visual effect. I mean, same Yeshwanth Reddy Yerraguntla: Okay. Yeah. Feedback feedback. Manisha Gundapuneedi: Uh, Yeshwanth Reddy Yerraguntla: I think everything is fine. Manisha Gundapuneedi: fine. Okay. status. Yeshwanth Reddy Yerraguntla: Okay, got it. Rajashekar G: Okay. One, two, Yeshwanth Reddy Yerraguntla: Mhm. Rajashekar G: back. Satyasri Prabhakar Mantripragada: I suggesting them to go by the plan P 0 P1 P2 we can execute that first break down those into them. Yeshwanth Reddy Yerraguntla: He asks I Satyasri Prabhakar Mantripragada: So we are making progress any project team there should be a Yeshwanth Reddy Yerraguntla: don't Satyasri Prabhakar Mantripragada: story that story should have tasks on how to accomplish the story. Yeshwanth Reddy Yerraguntla: specific Rajashekar G: Uh not only like every Click whatever we are doing. Satyasri Prabhakar Mantripragada: It is like a product Yeshwanth Reddy Yerraguntla: Right. Satyasri Prabhakar Mantripragada: and 00:10:21 Rajashekar G: But the point is that it will be easy to track Satyasri Prabhakar Mantripragada: task is log. It's not it's the other way around. Rajashekar G: him. Satyasri Prabhakar Mantripragada: It is that you know what is the task that you need to pick up. Yeshwanth Reddy Yerraguntla: He Satyasri Prabhakar Mantripragada: So task you are blocked you should be able to pick up the next Yeshwanth Reddy Yerraguntla: couldn't. Satyasri Prabhakar Mantripragada: task. So you have your plan for the next one or two weeks at least. That is how we have to progress. Yeshwanth Reddy Yerraguntla: You got issue process issue. Um Satyasri Prabhakar Mantripragada: Uh Manisha Gundapuneedi: drill down. Sorry, sorry. Satyasri Prabhakar Mantripragada: Sorry. Manisha Gundapuneedi: estimation like Yeshwanth Reddy Yerraguntla: drill Manisha Gundapuneedi: each. Yeshwanth Reddy Yerraguntla: down. Rajashekar G: Chan, chi Manisha Gundapuneedi: Ah, they Yeshwanth Reddy Yerraguntla: Meaning Rajashekar G: plus. Manisha Gundapuneedi: Monday and press on. Yeshwanth Reddy Yerraguntla: enterprise brain next four days we want to sit and come up with the whole list of tasks obviously I understand Satyasri Prabhakar Mantripragada: Oh, Yeshwanth Reddy Yerraguntla: essentially points what we have to tell them or what we are expecting for from them for this one week. 00:12:36 Satyasri Prabhakar Mantripragada: Yeah. Okay. Our objective is to create the platform. Yeshwanth Reddy Yerraguntla: Yeah. Satyasri Prabhakar Mantripragada: We are working Yeshwanth Reddy Yerraguntla: Yeah. Satyasri Prabhakar Mantripragada: on only At least for me, it is not Yeshwanth Reddy Yerraguntla: It is not clear. Satyasri Prabhakar Mantripragada: clear. Yeshwanth Reddy Yerraguntla: I agree. I completely agree with that. So sheet open Rajashekar G: Yeah, Manisha Gundapuneedi: That's okay. Yeshwanth Reddy Yerraguntla: we came up with these kinds of responsibilities and actual system architecture at least. What are we trying to Rajashekar G: I'm opening a good idea. What's happening? Yeshwanth Reddy Yerraguntla: Yeah, in the detail. Rajashekar G: I Yeshwanth Reddy Yerraguntla: So main we want to deliver enterprise brain in whatever state it is right Rajashekar G: don't Yeshwanth Reddy Yerraguntla: now to the extent that it is usable and after that we want to focus mainly on V2 architecture because V1 Rajashekar G: think Yeshwanth Reddy Yerraguntla: challen um Satyasri Prabhakar Mantripragada: Yeah. The the only thing is that they say yeah get started Yeshwanth Reddy Yerraguntla: ground 00:14:15 Satyasri Prabhakar Mantripragada: and we committed for I think two or 3 weeks of uh deliverable time. Yeshwanth Reddy Yerraguntla: 6 weeks. Satyasri Prabhakar Mantripragada: So okay Yeshwanth Reddy Yerraguntla: Yeah. Satyasri Prabhakar Mantripragada: max can be Yeshwanth Reddy Yerraguntla: Okay, got it. Satyasri Prabhakar Mantripragada: for development for Yeshwanth Reddy Yerraguntla: Got Satyasri Prabhakar Mantripragada: us to be done saying that Yeshwanth Reddy Yerraguntla: it. Satyasri Prabhakar Mantripragada: okay could be or Okay. Huh. Manisha Gundapuneedi: First version 6 weeks platform Satyasri Prabhakar Mantripragada: Okay. Okay. Manisha Gundapuneedi: week. Okay. Rajashekar G: Six weeks. Yeshwanth Reddy Yerraguntla: Six week Rajashekar G: Six weeks including plan for one week. Yeshwanth Reddy Yerraguntla: six week including Rajashekar G: Okay. Okay. Okay. Yeshwanth Reddy Yerraguntla: Right. So, we want to accomplish given the number of features proactive right that is that is the whole p Satyasri Prabhakar Mantripragada: Oh, Yeshwanth Reddy Yerraguntla: around which we are now moving deployment whatever we are Satyasri Prabhakar Mantripragada: good. Yeshwanth Reddy Yerraguntla: giving reactive right so proactive whatever you want to First system architecture top to bottom whether it takes 6 weeks or whether it takes 6 months it doesn't matter top to bottom even if it takes till December it's not what we should take shortcut at least show the show the whole list then we can decide at the end of six weeks how we can intelligently come up with the right set of tasks so that we can at least deliver something that they can use 500 00:16:56 Yeshwanth Reddy Yerraguntla: tasks till December. What are those 50 tasks which we can accomplish by 6 weeks so that to whatever extent we can deliver we can deliver. Um, am I making sense? Hello, Rajashekar G: Yes, Yeshwanth Reddy Yerraguntla: sir. Rajashekar G: sir. Satyasri Prabhakar Mantripragada: So at least yeah what he saying is the intention is that you know what you need to Yeshwanth Reddy Yerraguntla: Mute. Satyasri Prabhakar Mantripragada: develop Yeshwanth Reddy Yerraguntla: Correct. Exactly. Correct. Correct. Satyasri Prabhakar Mantripragada: hypo that okay that that's Yeshwanth Reddy Yerraguntla: First version. Satyasri Prabhakar Mantripragada: the Yeshwanth Reddy Yerraguntla: I'm not happy. So you put main two conraints sir um come up with this 500 list of items. Second is until you come up with that list of items what are our juniors Satyasri Prabhakar Mantripragada: Oh. Yeshwanth Reddy Yerraguntla: doing after that you will Satyasri Prabhakar Mantripragada: Oh. Yeshwanth Reddy Yerraguntla: be available to go do the actual architecture and planning and all those things on jetparu. So it's 6:30 1 00:18:30 Satyasri Prabhakar Mantripragada: Okay. Sorry. Sorry. Can you repeat the last one? Yeshwanth Reddy Yerraguntla: Once I have the full plan sorry once I can tell you what are all our juniors doing for this one week at least this one week I am freed with Manisha and Rajar to come up with the actual list of items Satyasri Prabhakar Mantripragada: Okay. Okay. Okay. I'm doing fine. Fine. Then Yeshwanth Reddy Yerraguntla: Ganesh they have to work on the Tata steel side of it. Tatum data Satyasri Prabhakar Mantripragada: Mhm. Yeshwanth Reddy Yerraguntla: connectors I don't know how many connectors are going to come now Manisha Gundapuneedi: Foreign speech. Yeshwanth Reddy Yerraguntla: unify. Manisha Gundapuneedi: Foreign speech. Foreign speech. Yeshwanth Reddy Yerraguntla: Yeah. To whatever extent Like we can debate debate on the rest. Manisha Gundapuneedi: Okay. Amen. foreign. Yeshwanth Reddy Yerraguntla: Yeah. Manisha Gundapuneedi: So time Yeshwanth Reddy Yerraguntla: Okay. Well, research Manisha Gundapuneedi: and Yeshwanth Reddy Yerraguntla: and 00:21:12 Manisha Gundapuneedi: timeline. Yeshwanth Reddy Yerraguntla: Friday. Um and we harsh a show Manisha Gundapuneedi: I'm Rajashekar G: Amoleia. Manisha Gundapuneedi: here. Yeshwanth Reddy Yerraguntla: probably Rajashekar G: Okay. So, Yeshwanth Reddy Yerraguntla: then we'll be free. Rajashekar G: so first thing Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: about Yeshwanth Reddy Yerraguntla: Plot graphs outl That's a lot. Rajashekar G: manufacturing to for enterprise. Enterprise Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: Republic. Yeshwanth Reddy Yerraguntla: Enterprise brain. Rajashekar G: Fork Yeshwanth Reddy Yerraguntla: branch manufacturing Rajashekar G: for Yeshwanth Reddy Yerraguntla: demo manufacturing brain actual dashboard Rajashekar G: I'm sharing this Yeshwanth Reddy Yerraguntla: and a half or two hours slow. We'll get back to you. Satyasri Prabhakar Mantripragada: Sure that we also pull Yeshwanth Reddy Yerraguntla: Mainly the water s Satyasri Prabhakar Mantripragada: in Yeshwanth Reddy Yerraguntla: okay. Satyasri Prabhakar Mantripragada: you saying something. Yeshwanth Reddy Yerraguntla: What are the task for this week? Uh, next week only architecture Satyasri Prabhakar Mantripragada: Yes, Yeshwanth Reddy Yerraguntla: focus. Satyasri Prabhakar Mantripragada: if you are spending time spend it for next two weeks so that you'll have some bandwidth to plan for next two weeks only one week. 00:24:02 Yeshwanth Reddy Yerraguntla: Okay. Okay. Okay. Okay. Okay. Set up. Okay. Okay. Sorry. Sir, noted. Rajashekar G: Bye, sir. So now what I'm saying One more thing. Yeshwanth Reddy Yerraguntla: Yeah. Oh no. Oh no. Rajashekar G: It will work. Not an issue. Yeshwanth Reddy Yerraguntla: Like I speaking, enterprise brain tomorrow. Rajashekar G: M. Yeshwanth Reddy Yerraguntla: So ideally enterprise it could be its own branch. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: Yeah. Integration. Yeah. Sorry. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: Okay, Rajashekar G: and Yeshwanth Reddy Yerraguntla: just one minute different topic Rajashekar G: so Yeshwanth Reddy Yerraguntla: open. I'm testing Pantic AI gateway. Rajashekar G: open Yeshwanth Reddy Yerraguntla: Okay. $5. Rajashekar G: access. Yeshwanth Reddy Yerraguntla: Sir, Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: we can continue this discussion. Rajashekar G: Yeah. Yeshwanth Reddy Yerraguntla: Huh? Screen share. Yes. 00:27:36 Yeshwanth Reddy Yerraguntla: Know I know sheets dark mode. Next two weeks. Most connectors for Manisha Gundapuneedi: However, Yeshwanth Reddy Yerraguntla: Discord Manisha Gundapuneedi: there Yeshwanth Reddy Yerraguntla: already. Uh connector generic rest API connector. Manisha Gundapuneedi: and Yeshwanth Reddy Yerraguntla: Okay. Manisha Gundapuneedi: already Yeshwanth Reddy Yerraguntla: Next Are we on leave? Manisha Gundapuneedi: next week. Yeshwanth Reddy Yerraguntla: Okay. Next week already I would suggest he take up uh skills run books because he has been working on system prompt side of enterprise brain. Amoleia on the other hand can take up frag starting responsibility. Manisha Gundapuneedi: First for Yeshwanth Reddy Yerraguntla: connector Manisha Gundapuneedi: type. Yeshwanth Reddy Yerraguntla: creatity. Two weeks Manisha Gundapuneedi: Ask Yeshwanth Reddy Yerraguntla: already user So Manisha Gundapuneedi: money. Yeshwanth Reddy Yerraguntla: user I need Manisha Gundapuneedi: Okay. Yeshwanth Reddy Yerraguntla: true. That is true. Yeah, roughly level resolution local we'll have some Rajashekar G: Sure. Yeshwanth Reddy Yerraguntla: clarity. So we have three components. 00:32:42 Yeshwanth Reddy Yerraguntla: Um half an hour just Manisha Gundapuneedi: Okay. Rajashekar G: You don't understand. Manisha Gundapuneedi: Data. Yeshwanth Reddy Yerraguntla: Uh yeah, Manisha Gundapuneedi: makeup. Yeshwanth Reddy Yerraguntla: technically Rajashekar G: Okay, let's write down the points Manisha Gundapuneedi: Okay. Rajashekar G: on Yeshwanth Reddy Yerraguntla: Uh yeah I mean but we have to come up with at least high level stories. Manisha Gundapuneedi: For Rajashekar G: Right. Manisha Gundapuneedi: example, for Rajashekar G: Okay. Okay. Yeshwanth Reddy Yerraguntla: True. But you can try Rajashekar G: I'll write few points. Yeshwanth Reddy Yerraguntla: Yeah. Exact. Manisha Gundapuneedi: 8:15. Yeshwanth Reddy Yerraguntla: We'll unmute and Manisha Gundapuneedi: Okay. Rajashekar G: sheet. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: Done. Now it's 7:15. Manisha Gundapuneedi: platform. Yeshwanth Reddy Yerraguntla: Yeah. 5 minutes. Rajashekar G: Fine. We can connect Yeshwanth Reddy Yerraguntla: time. Rajashekar G: screen. Yeshwanth Reddy Yerraguntla: Got it. Rajashekar G: So I break down into these three parts. Okay. Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: Real time intelligence knowledge based related work. 01:05:07 Rajashekar G: Third one is this refresh. So connection Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: establishment. Yeshwanth Reddy Yerraguntla: Okay. First thing real Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: agent user awareness whole list Rajashekar G: volume, Yeshwanth Reddy Yerraguntla: both are different agents. You agree right? Rajashekar G: right? Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: I think I'm thinking Yeshwanth Reddy Yerraguntla: H. Rajashekar G: data source agent. Yeshwanth Reddy Yerraguntla: Oh Rajashekar G: But this Yeshwanth Reddy Yerraguntla: no. Rajashekar G: again only user user awareness. Yeshwanth Reddy Yerraguntla: data agent in data snapshot data. It is synthesizing and telling what is the what are all the important uh news items. Rajashekar G: So user key always info. Yeshwanth Reddy Yerraguntla: No waiting complete those are first it is synthesizing the news items. Okay. Job one synthesize news. Job two update knowledge graph. Job three, news items n for every user give the necessary information on job three. If if we go in the route of getting rid of that blue box blue or green or 01:07:08 Rajashekar G: This Yeshwanth Reddy Yerraguntla: whatever that third job user aware Rajashekar G: one Yeshwanth Reddy Yerraguntla: intelligence if we get get rid of that green box job three will fall on the red box. Rajashekar G: job 3 minutes. This Yeshwanth Reddy Yerraguntla: What job three is uh for every member I have to tell Rajashekar G: one, Yeshwanth Reddy Yerraguntla: them news in their own fashion just the job three cost money Rajashekar G: right? Yeshwanth Reddy Yerraguntla: blue box you know because for every user if I have to decide what is important what is not important then I felt it is too much burden on the red box red box will simply say okay this is news someone take care of for every user what needs to be broadcasted that itself is a very large task on Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: pitch if there are 100 users how smartly it has to figure out right so Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: already to fetch the snapshot to understand what is going on what is new that itself is a large task and picture. You can 01:08:26 Rajashekar G: So user key updated this one like whenever user asks a Yeshwanth Reddy Yerraguntla: counter Rajashekar G: query so there will be single communication agent that needs to be providing the answer let's say user question okay interface user question okay but There is like this distribution Yeshwanth Reddy Yerraguntla: kind of. Yeah. Rajashekar G: and these two also in sync with the information in this Yeshwanth Reddy Yerraguntla: Which two? Rajashekar G: agent and this agent. Yeshwanth Reddy Yerraguntla: Yes. Rajashekar G: Okay. Now this will break. We can break this into multiple Yeshwanth Reddy Yerraguntla: And from the red box, Rajashekar G: things. Yeshwanth Reddy Yerraguntla: there are still two pathways to the main user. Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: In the form of direct question answering and in the form of critical Rajashekar G: Mhm. Yeshwanth Reddy Yerraguntla: alerts. Something is really critical. H actually critical critical user awareness names box one box whatever give Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: some already name per Rajashekar G: One, two. Yeshwanth Reddy Yerraguntla: 3. Rajashekar G: This is will be 01:10:36 Yeshwanth Reddy Yerraguntla: 46. Yeah. I mean to agree on what are the roles of 1 2 3 4 5 6 they itself is a big Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: good Rajashekar G: Asana if uh first of all if you dive into the architecture Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: also this is a pentic architecture set up if it is a micros service we are going to build like we will be able to break through and reach all the points and Yeshwanth Reddy Yerraguntla: Yeah, Rajashekar G: picture. Yeshwanth Reddy Yerraguntla: true. So, Box number Rajashekar G: One key Yeshwanth Reddy Yerraguntla: four. Rajashekar G: always. A minor. Okay. Yeshwanth Reddy Yerraguntla: See whatever is static knowledge whatever is dynamic knowledge I don't want to keep system prompt in a string in python Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: similar tools also you must have heard of this term skills, right? Rajashekar G: Yes. Yeshwanth Reddy Yerraguntla: Skills are nothing but tools. Rajashekar G: But what I'm thinking is tools like Salesforce. 01:13:12 Rajashekar G: For example, Yeshwanth Reddy Yerraguntla: Mhm. Here I can in our case Let's not talk about data connector related anything because those are data connector related info. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: In our case when we are dealing with realtime decision intelligence and user aware notification system the corresponding skills are going to be like enterprise brain low skill. skills. Whenever I'm thinking of skills, I'm thinking of run books only. Rajashekar G: both are different and this like runbook sir whenever uh we are retraining something it it already know Yeshwanth Reddy Yerraguntla: Run books are mainly processes, processoriented uh knowledge. Rajashekar G: means like user user Yeshwanth Reddy Yerraguntla: Ku not I can give you one runbook which is like when you're onboarding Rajashekar G: training Yeshwanth Reddy Yerraguntla: a fresh project here are all the 20 steps that you have to do. First step create a lead in Salesforce. Second step create a group chat in uh Google chat. Third step create a share drive in Google Drive. Fourth step assign a PM. 01:14:51 Yeshwanth Reddy Yerraguntla: Assign a product owner. Uh create a fresh GitHub repo is the runbook based on the runbook and Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: knowledge organization snapshot because in the snapshot it will know right is a GitHub repo created or not is a where is the Salesforce lead generated? Who is the product owner? Rajashekar G: Thank you. All Yeshwanth Reddy Yerraguntla: Based on certain things available, certain things not available it will know. Okay, in the 20 steps only five have been done so far. What does it take to do the next five steps? I have to notify these five people. I have to create a task uh kind of a thing on their Rajashekar G: right. Yeshwanth Reddy Yerraguntla: dashboard. I have to notify that this is pending on uh so and so person to their managers. This is run book. Rajashekar G: You're Yeshwanth Reddy Yerraguntla: Skilled on the other hand is like um how to update Rajashekar G: right. Yeshwanth Reddy Yerraguntla: snapshot how to how to understand user's intent and convert it to uh user preferences in our database. 01:16:08 Yeshwanth Reddy Yerraguntla: There is no process involved. Rajashekar G: All right. Yeshwanth Reddy Yerraguntla: Okay. Runbook is like blindly you do these 20 steps. So if there are anything pending you take care of how to handle escalations. Runbook is uh skill is like on the spot what is the best thing to do how to do it efficiently. Rajashekar G: And there skill is nothing but system prompt. Yeshwanth Reddy Yerraguntla: Skill is multiple system prompts. Rajashekar G: What to do with what to do with which Yeshwanth Reddy Yerraguntla: Multiple Rajashekar G: data? Yeshwanth Reddy Yerraguntla: correct system prompt will always be lightweight. You are core agent. Your job is to fetch um regular information. Your job is to put it in snapshot. Your job is to get snapshot and uh broadcast news. the system prompt. Now when a chat comes saying um hey I'm a trainer uh please add this new step into uh this so and so run book. So what to do when some trainer is asking me to update the run book 01:17:34 Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: in a sense skill is like a tool that Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: um it will discover those tools on the fly. Rajashekar G: on the flip tools. Discord tools. We will be writing Yeshwanth Reddy Yerraguntla: H we will write the tool but it will never be in system Rajashekar G: down Yeshwanth Reddy Yerraguntla: prompt. You'll never write it like um at the rate agent tools Rajashekar G: Then how it will figure out like uh you have these tools. Yeshwanth Reddy Yerraguntla: to know that it will first we will have all those tools in in their own Python files. Okay. First thing we never do agent tool add the rate but what we'll do is Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: we'll say at the rate agent dot skill folder Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: folder low how many of our tools are there they are all how many our functions are there they are all potentially tools but it will never discover those Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: tools the load system that is how you save a lot of memory and 01:18:50 Rajashekar G: H Yeshwanth Reddy Yerraguntla: time. One homework I'll suggest is just go through how cloud skills work. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: You will understand immediately what I'm trying to say. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: So what are the tools? Run query show visualization rack cache here show visualization is not the job of core agent right Rajashekar G: I did someone Yeshwanth Reddy Yerraguntla: UI agent UI Rajashekar G: agent. Yeshwanth Reddy Yerraguntla: agent the purple box Rajashekar G: This one Yeshwanth Reddy Yerraguntla: right ah user interface is slow Visualization intelligence agent Rajashekar G: H. Yeshwanth Reddy Yerraguntla: box number two. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: Yes. Rajashekar G: Mhm. Yeshwanth Reddy Yerraguntla: See uh every agent should do only one thing. Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: If you make it do too many things um chances of failure will constantly increase. Rajashekar G: Right. agent. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: This one will be continuously monitoring all this. Yeshwanth Reddy Yerraguntla: And let me just Rajashekar G: It also in communication always and whatever going Yeshwanth Reddy Yerraguntla: Yes. Rajashekar G: upward user it will go through another agent not directly to the UI. 01:21:18 Yeshwanth Reddy Yerraguntla: H correct. Whatever has to go in on the UI is decided by the visualization Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: agent. Rajashekar G: Try it down. There you go. So overall 3 4 Five major Yeshwanth Reddy Yerraguntla: I mean refresh agent need not be called an Rajashekar G: concrete that's a Yeshwanth Reddy Yerraguntla: agent. Rajashekar G: refresh mechanism Yeshwanth Reddy Yerraguntla: H. Rajashekar G: mechanism. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: So refresh mechanism that will be a trigger for this agent. Yeshwanth Reddy Yerraguntla: Yes. Rajashekar G: It is as simple as like uh at 6:00 6:00 we are asking these questions to this agent Yeshwanth Reddy Yerraguntla: No. No. Rajashekar G: and Yeshwanth Reddy Yerraguntla: It's 6:00. These are all We are currently connected to 20 data sources. Of those 20, please refresh these five data Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: sources. Rajashekar G: Common scheduleuler will be there. Yeshwanth Reddy Yerraguntla: Yeah, Rajashekar G: A configuration file for setting up when the refresh should trigger for each data agent. Yeshwanth Reddy Yerraguntla: correct. Rajashekar G: How will the agent know which data should fetch while refresh from the data data source agent first 01:24:15 Yeshwanth Reddy Yerraguntla: Correct. Rajashekar G: info Yeshwanth Reddy Yerraguntla: How will the agent know which data should fetch refresh from data source dat 1 minute this refresh mechanism is fetching from posgress uh what are all the agents to Rajashekar G: Mhm. Yeshwanth Reddy Yerraguntla: refresh the moment it realizes So of these 20 these five have to be refreshed. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: It will send a message to realtime intelligence agent. This is a purely time based u u activity. Every minute uh refresh mechanism is uh deciding should I send, should I send, should I send, should I send or not. Rajashekar G: Right. What I'll try to do dig deeper into all these in depth. Yeshwanth Reddy Yerraguntla: Mhm. Rajashekar G: I'll try to write the points and descriptions as what exactly needs to be done at Yeshwanth Reddy Yerraguntla: Okay. Rajashekar G: that particular point or not. Yeshwanth Reddy Yerraguntla: So, I'll share my screen now and I'll tell what I have uh discovered in the last Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: half an hour. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: It will it might help it know just first thing is um I gave ch the architecture just a screenshot and said um 01:25:46 Rajashekar G: Can Yeshwanth Reddy Yerraguntla: I want to create a system design LLD for this HLD HL. Rajashekar G: that gives me boatloads of information. Yeshwanth Reddy Yerraguntla: Yeah. So it has access to skills. So I brought a skill called uh decomposition reconstruction. It's a system design skill. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: It said first of all you have these six modules. Six. This is not a module core agent. Just five modules. 1 2 3 4 5 to some Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: extent roughly these are the six things that we have to build. Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: Next what it said is in each irrespective of the modules there are work streams. Um so first thing is context propagation and from chat interfa from dashboard to chat interface to core how the message is passed right and back answer so um I am trying to capture all the packets Rajashekar G: Hello. Yeshwanth Reddy Yerraguntla: um that are traveling packets on data whatever here. 01:27:21 Yeshwanth Reddy Yerraguntla: One packet is add queries followed by uh Q answers packet is uh dashboard to uh uh chat interface information passing and which these are these are the main data packets that are moving. Uh Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: second next life cycle. This is all being called as a life cycle. life cycle. Next, orchestration and answer life cycle. Uh, sorry. Yeah, Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: sorry. Sorry. Context propagation dashboard to chat. Chat to Mhm. Yeah, my bad. This is the only traveling packet information again. This goes to core Rajashekar G: What? Yeshwanth Reddy Yerraguntla: agent. User info should also be passed to core Rajashekar G: Right Yeshwanth Reddy Yerraguntla: agent. So who is asking a name? Chat UI agent to core. Rajashekar G: out Yeshwanth Reddy Yerraguntla: This is the core message orchestration and answer life cycle. So we are dealing with ad hoc queries to answers. 01:29:30 Rajashekar G: and training lessons. Yeshwanth Reddy Yerraguntla: Yes, exactly. Uh training lesson separate lessons chat UI agent core. training lessons separate. This is basically par queries answers. So as well as uh dashboard uh to sorry news to dashboard. Visualization intelligence knowledge life cycle life cycle flow. This is not knowledge life cycle. This is just uh core notifications user alert life cycle traveling packets. uh critical messages. So mainly you're talking about what are the broad highle packets that are moving around you will end up with something called as Rajashekar G: All right. Yeshwanth Reddy Yerraguntla: uh um sequence diagram. Right. So then at every point you will now Rajashekar G: I Yeshwanth Reddy Yerraguntla: know exactly what information to send. What is this an API request? Is this a database call? We clarity when you know it's an API request then you ask the Rajashekar G: got you. 01:31:49 Yeshwanth Reddy Yerraguntla: question what is the uh packet API Rajashekar G: I got Yeshwanth Reddy Yerraguntla: packet. So slow slowly step by step step by step you'll have more Rajashekar G: it. Yeshwanth Reddy Yerraguntla: details right that said um uh can we bypass and say okay next two weeks what will do or what will do answers on the other Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: hand here the way you have created the EV questions they are jumping into low level immediately. Rajashekar G: Mhm. Yeshwanth Reddy Yerraguntla: to uh the way we will probably come up with task is okay this is okay I'll ask to now create a database and put up this Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: task Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: we don't know how many questions are the right number of questions Rajashekar G: Google.com. Yeah. Yeshwanth Reddy Yerraguntla: I'm saying are these sufficiently enough number of questions. Rajashekar G: No, in Yeshwanth Reddy Yerraguntla: So a clarity structure this is one diagram. Rajashekar G: Got it. Yeshwanth Reddy Yerraguntla: One diagram gave rise to let's say seven rows. Each row will give rise to one diagram in sequence and in in that sequence you'll have like at least seven 01:33:19 Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: eight interactions. Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: So seven rows each row will have a sequence diagram of 74 you'll have 49 questions immediately there to answer all yeah so I'll Rajashekar G: Okay, Yeshwanth Reddy Yerraguntla: continue working on this no I'll suggest you retry I Rajashekar G: sure. Yeshwanth Reddy Yerraguntla: s we'll drop off for the time Okay, Rajashekar G: If something comes Yeshwanth Reddy Yerraguntla: sir. Rajashekar G: up, Satyasri Prabhakar Mantripragada: Salesforce Yeshwanth Reddy Yerraguntla: Concent. Hello. Satyasri Prabhakar Mantripragada: implementation for the Yeshwanth Reddy Yerraguntla: Huh? Satyasri Prabhakar Mantripragada: integration or getting some business knowledge and all Yeshwanth Reddy Yerraguntla: Like if you want to build a demo for you. Satyasri Prabhakar Mantripragada: anything in understanding to understand Salesforce. We should know what kind of objects to be quered implementation standpoint technical we can Yeshwanth Reddy Yerraguntla: Salesforce we ma when when a company creates a Salesforce connector Satyasri Prabhakar Mantripragada: all Yeshwanth Reddy Yerraguntla: um along with the tables will there be any catalog or any additional information that we can consume to teach the AI. What is what 01:35:14 Satyasri Prabhakar Mantripragada: you will have to ask Yeshwanth Reddy Yerraguntla: currently Satyasri Prabhakar Mantripragada: I'm not able to go Rajashekar G: Okay, Satyasri Prabhakar Mantripragada: down Rajashekar G: sir. Can you continue? Satyasri Prabhakar Mantripragada: and it can be not just for it can be for any future customer if they want integration Yeshwanth Reddy Yerraguntla: H yeah I mean Rajar if you see my screen basically what Satyasri Prabhakar Mantripragada: for Yeshwanth Reddy Yerraguntla: does it take to build a Salesforce connector question. So if we can see the screen concepts world model snapshot escalation. Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: So whatever questions we have to ask they should come in this uh location only. Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: So where is the organization knowledge? Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: Will the connector? Rajashekar G: Repeat point Yeshwanth Reddy Yerraguntla: Okay, sorry. Rajashekar G: repeat. Yeshwanth Reddy Yerraguntla: When there is a Salesforce connector, will it also come with a catalog? Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: Because catalog is what we are calling as org knowledge. Rajashekar G: So is nothing but then table Yeshwanth Reddy Yerraguntla: Sorry sir. 01:36:44 Rajashekar G: schema. Yeshwanth Reddy Yerraguntla: Our knowledge our knowledge is table schema data resources is cataloges. Rajashekar G: Okay. Satyasri Prabhakar Mantripragada: One other aspect that we need to consider is uh if we implement ARBback or some kind of access controls, how do they overlap or override Salesforce's implementation or whichever gets honored? Rajashekar G: And one more point we may get to know like if they're directly exposing the Salesforce. Yeshwanth Reddy Yerraguntla: Einstein. Rajashekar G: So API Yeshwanth Reddy Yerraguntla: Yes. Rajashekar G: direct Yeshwanth Reddy Yerraguntla: Salesforce Rajashekar G: H. Yeshwanth Reddy Yerraguntla: um table schema Satyasri Prabhakar Mantripragada: object. Yeshwanth Reddy Yerraguntla: data resources objects Custom objects. First question when I said when I use the word catalog that is basically trying to understand how Satyasri Prabhakar Mantripragada: Mhm. Yeshwanth Reddy Yerraguntla: to read and understand custom objects and custom Satyasri Prabhakar Mantripragada: Should Yeshwanth Reddy Yerraguntla: columns. Satyasri Prabhakar Mantripragada: we know the terminology? Uh what all entities can exist within the context of sales or we can train them. Yeshwanth Reddy Yerraguntla: Yeah. So true. True. We need to know what are the custom terminologies that Salesforce uses and organization law when every organization will have its own pipeline uh configured. 01:39:22 Yeshwanth Reddy Yerraguntla: Uh some organizations will say first leader then something then maybe in a different Rajashekar G: Right. Satyasri Prabhakar Mantripragada: What? Yeshwanth Reddy Yerraguntla: organization there's no lead at all. Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: So where are those uh pipeline uh related informations technically but you can still uh it's worth Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: asking. Rajashekar G: Right now whatever we have done simple Salesforce it will use the Salesforce Yeshwanth Reddy Yerraguntla: Oh no. Rajashekar G: API. So we haven't dig deep into like Salesforce Yeshwanth Reddy Yerraguntla: Correct. Rajashekar G: schema is taking the basy natural language agent is doing that. But Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: here Yeshwanth Reddy Yerraguntla: Custom fields. Um, Rajashekar G: right it's working. If someone custom people Yeshwanth Reddy Yerraguntla: yes. Rajashekar G: comes Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: login. Yeshwanth Reddy Yerraguntla: But there will be a lot more in the back. But I completely understand your Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: question. Unknown columns. How will the agent know? 01:41:32 Rajashekar G: Okay. kind of. But first of all, Yeshwanth Reddy Yerraguntla: Yeah. Can we call Salesforce uh custom Salesforce agents or Satyasri Prabhakar Mantripragada: Thank you. Yeshwanth Reddy Yerraguntla: intelligence? Einstein. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: When the time comes to meeting screenshot, you try to convert it into Salesforce terminology. So when we talking about snapshot, right? What does it mean in Salesforce terminology to have a snapshot? Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: What is the current uh uh state like for every uh company where it is is the snapshot? Snapshot uh every company different account uh so then the question becomes uh I want to understand for every company what is the uh process they follow for updating their uh leads or updating their pipelines. That is the question you'll ask once and you'll simply ask um because every company will be unique in their own pipelines. So how can I find that information on a company to company basis? Right? 01:43:22 Yeshwanth Reddy Yerraguntla: Just reading that snapshot we I was able to come up with that question. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: Next escalation rule zoom like if something goes wrong usually one of our enterprise brains feature is we have to escalate it to the right person. For every company this might change again when a lead simply vanishes. So for one company it might mean nothing but for one other company it might be very severe. Rajashekar G: Right. Yeshwanth Reddy Yerraguntla: So where can I find that information for a company to company basis? Is it something that um is present in in a particular standardized location or do we have to capture it uh all the time? So next you might go through runbooks blah blah blah and try to come up with questions on the fly. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: When the moment you are able to frame those questions, it also means you are able to understand how data agent itself works. That is my um intention. Rajashekar G: Okay. So this data source agent can be done by Manisha's team. 01:44:26 Yeshwanth Reddy Yerraguntla: Actually Satyasri Prabhakar Mantripragada: Okay, Rajashekar G: So if they are also in the part Satyasri Prabhakar Mantripragada: sure. Yeshwanth Reddy Yerraguntla: 15 minutes go through this one block you yourself will have Rajashekar G: Sure. Yeshwanth Reddy Yerraguntla: questions. Yeah. Rajashekar G: And the Salesforce connect. Satyasri Prabhakar Mantripragada: specialist. Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: Concent. Rajashekar G: Okay. Satyasri Prabhakar Mantripragada: Every company decide Let's say Rajashekar G: Okay. Satyasri Prabhakar Mantripragada: Dr. Rajashekar G: Okay. Satyasri Prabhakar Mantripragada: Okay, fine. Rajashekar G: Lind. Yeshwanth Reddy Yerraguntla: H Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: I mean available. Satyasri Prabhakar Mantripragada: Iv Yeshwanth Reddy Yerraguntla: Yeah. Okay. Rajashekar G: Hey Yeshwanth Reddy Yerraguntla: I think you know committee Satyasri Prabhakar Mantripragada: time. Yeshwanth Reddy Yerraguntla: sessar Satyasri Prabhakar Mantripragada: Look. Yeah. Right. Yeshwanth Reddy Yerraguntla: Yeah, they're free now. Okay. Done. And Satyasri Prabhakar Mantripragada: Yeah. Thank you. Yeshwanth Reddy Yerraguntla: connect. Okay. Manisha Gundapuneedi: Evening. Yeshwanth Reddy Yerraguntla: Sir now whatever whatever you are able to do um whenever you are finishing Manisha Gundapuneedi: Okay. Yeshwanth Reddy Yerraguntla: yeah Rajashekar G: Okay. Yeshwanth Reddy Yerraguntla: tablec See two weeks. So yeah, we can probably say what is the MVP we can deliver in each box that we can deliver in two weeks. Right. Manisha Gundapuneedi: Okay. Yeshwanth Reddy Yerraguntla: Yeah. Rajashekar G: Clear while doing there might be points Yeshwanth Reddy Yerraguntla: Sir. Rajashekar G: of different Yeshwanth Reddy Yerraguntla: Yeah. Sorry. Manisha Gundapuneedi: Okay, I'm dead. Rajashekar G: manufacturing deployment. Manisha Gundapuneedi: Bye. Rajashekar G: Okay. Satyasri Prabhakar Mantripragada: Okay, Rajashekar G: Enterprise for manufacturing Satyasri Prabhakar Mantripragada: Okay. Rajashekar G: complete office. Satyasri Prabhakar Mantripragada: Sure. Rajashekar G: Okay. Satyasri Prabhakar Mantripragada: Okay. Rajashekar G: will be working. Satyasri Prabhakar Mantripragada: Okay. Uhhuh. Rajashekar G: Okay. Okay. Oh, now just by Transcription ended after 01:50:45 This editable transcript was computer generated and might contain errors. People can also change the text after it was created.