Feb 23, 2026 Meeting Feb 23, 2026 at 04:53 PST - Transcript 00:00:00 Rajashekar G: Hi guys. recording is enabled now. So guys, Rahul, are you able to hear me? Hello Rahul. Can you hear me? Rahul Mishra: Hello. Rajashekar G: Yeah, I'm sharing my screen. So, Rahul Mishra: Yes. Rajashekar G: so enterprise brain I'll start from here. So actually uh whatever you have done in optin so enterprise brain is almost same thing. So enterprise brain is something like there will be a pentic agent. It will be having a lot of system form and there will be some connectors like Salesforce, Rahul Mishra: Okay. Rajashekar G: guit and insurance DB. The database is like the NLP to SQL processing one which one is used for IR database actually a replica of IR database. So this is having three connectors. Rahul Mishra: Okay. Rajashekar G: So right now where we are means like uh there are two environments one is enterprise brain. dwami.com so in which we are having only Salesforce intelligence as of now so we focused on this way why because most of the clients are uh were asking for Salesforce initially so this like uh only Salesforce intelligence is running in here 00:02:23 Rahul Mishra: H. Rajashekar G: another one is doo enterprise.com this is another environment here we'll be having like Salesforce and guit and insurance we are connected okay so what's happening inside the application is uh there is a pent agent having system count and the connectors and for NLP to SQL it will be having a tools like uh to generate SQL query I'll share the code also so this is I'll start with pyog so to achieve this entire thing where we started is using pyog actually so you guys already familiar with pyog which is a package developed by Um so it will be having like websockets identic agent uh instantiation and chat interface uh what we can say session management everything is there like tool calls how tool calls will happen how we need to debug and everything is already there. So we have this pileup on top of this we are having a another repository which is a enterprise brand. So in which we will be having this agent in agents folder let's say back end app agents EB executing agent. So actually there was a different architecture we planned like planning agent will be there executing agent will be there likewise but right now we ditched everything and we relied on one page. 00:03:58 Rajashekar G: This is agent.py. So in which we will be having this uh connectors. This is DB config. This is another one and insurance DB. And we'll be having tools like get DB connection, get app DB connection, get insurance DB connection, get Salesforce tables, load database schema, load insurance schema. Likewise, these are all the system prompt and we'll having like output how output need to be flow like this is log query to MD generating embedding. So what will happen is in guit one is we will be having another um we are having like uh two applications here to run like local host col 8000 and local host col 510 510 is running on uh using pyog completely. So in pyog we are using this agent.py PI file to do whatever like whatever we write here and we are having tools for generating uh charts generating graphs using a package called plotly that is also inside in pyog. So whenever any UI change or something comes we are making changes in uh pyog we forked another repository from pyog and we are making changes and we are merging this into pyog repository and pyog will be deployed I mean uh the new version will be updated that we will be using whenever any UI change comes any that kind of functionality comes like major core functionality changes comes another logic is on the wrapper whenever like guiting guit uh data embedding and a implementation tool modification let's say 00:05:42 Rajashekar G: um like charts chart scal like tools whatever we need like plotly rendering chart like alter chart all these will be written in enterprise plan okay so here what's happening is as of now this login and everything is also using pylog only let's say I'm asking a question show relationship between amount and probability. So these are the tasks happening visualizing amount probability counting opportunities checking custom opportunity schema visualizing revenue pro these are all tool calls happening and once this started completing the UA will be rendering something this is the approach as of now. So now regarding the governance and all what you people need to do is uh system prompt system prompt first part like as of now we are like directly exposing the system prompt in through logs someone will see the application that will be directly exposed that is one part how we will be like able to encrypt the system prompt income What is like I don't have the clarity. So as of now so first part is like go to enterprise make sure everything will be pushed into develop by today or tomorrow we'll do like uh as of now you guys want to start I'll share this branch in the chat go to this branch and do Sorry, I'll update the 00:07:32 Arpit Pathak: I didn't have a question. Rajashekar G: tell. Arpit Pathak: Okay. For example, uh when you say that we have to encrypt the system prompt, Rajashekar G: Sorry. Arpit Pathak: uh when you say that we have to encrypt the system prompt. So what does that actually mean to to the person who is using it to the logs or Rajashekar G: uh that I do I do don't have this entire clarity we need to see like Ganesh Arpit Pathak: where Rajashekar G: have like little bit clarity actually no same as like as we discussed on Friday even if we have the access to the database we won't be able to see the passwords right I mean even if we think we don't know the passwords right because they are encrypted same way if Someone has the database access or someone has the code access. Arpit Pathak: Yeah. Rajashekar G: They should not be able to see the system routes. Very well. Arpit Pathak: No code access if they have then they can see exactly Rajashekar G: Yeah. Arpit Pathak: right. Rajashekar G: You can see they have the token actually that actually even the frame passwords can also be seen if we have the token, right? 00:08:45 Rajashekar G: But in production actually won't be giving the tokens won't giving the encryption key people will going some Arpit Pathak: Uh, Rajashekar G: people are going to know actually um but uh Arpit Pathak: that's Rajashekar G: that's but where where we need to encrypt where we need to expose that clarity Arpit Pathak: all. Rajashekar G: we need to have yeah that that one the scope is not yet fixed so we need to come up with what can be possible or what cannot be possible. as well at the earlier stage. If we give those points to Raar or someone tomorrow and the first off itself then they will guide us like where to start and how to proceed. Good. Arpit Pathak: Mhm. Rajashekar G: So here is the main file. Arpit Pathak: Again I have still initial out. Rajashekar G: Yeah. Arpit Pathak: So let's say we have the system prompt in the file not in the code separate file. Rajashekar G: Okay. Arpit Pathak: uh and that is being encrypted and saved somewhere it's S3 okay now when we use the system prompt right would that stored system prompt be encrypted or we need to encrypt it later after storing 00:10:26 Rajashekar G: We need to encrypt while storing it. Uh after storing it, after retrieving we need to decrypt it, send it to AI actually. Yeah. So storing itself uh we need to encrypt it and store Arpit Pathak: Okay. Rajashekar G: it. Are you Arpit Pathak: Huh? Rajashekar G: clear? Arpit Pathak: No no no no. Uh so let's say if we are logging everything in uh this uh log fire we will see the system prompt Rajashekar G: Mhm. Yes. Arpit Pathak: right Rajashekar G: But that we should stop like wherever system prompt is exposing to the outer world Arpit Pathak: there. Rajashekar G: we need to stop it. Even though if we give the application to any uh like outsiders like if we like once we are giving this letter to the client also the client should not be able to read the ent Arpit Pathak: Huh? Rajashekar G: system from I guess somehow I think able to read directly he cannot able to read he need to decrypt it and read I think key should be there yeah Rahul Mishra: like 00:11:40 Rajashekar G: regarding that we'll discuss that the plan the plan is not yet finalized we need to Arpit Pathak: Okay. Rajashekar G: come up with the points and they'll also guide us like what are the things need to be Arpit Pathak: Okay. Rajashekar G: done okay so like once you've done like you Arpit Pathak: Mhm. Okay. Rajashekar G: will go to backend folder while running the application enterprise a enterprise don't clone it and go to back end folder in the back end uh you initiate uh VNV and you will you can do UV sync in env.de We need to paste the environments folder in the environments folder env and this is the command like I'll share this command in the chat we need to update the readme actually we'll update and uh keep it today itself creating activity and then so this command will run the application. That's it. I think one is uh you guys know pyog right how pyog Arpit Pathak: Mhm. Rajashekar G: works. So this is a pylog UI file executing agent pylog u. So this is the file we are running and the major enter 00:13:40 Arpit Pathak: Uh-huh. Rajashekar G: agent code will be here agent.py And the enter UI code is in I mean pyogi.py single page not enter pyog UI this is just to initiate this app factory and all. Okay. But everything will be placed inside the pyog library. This is the fork library actually. So we need to make changes to pog library. If any UI changes are like what I can say a change that need to be there forever. uh if you like migrate a change like uh what I can say sockets has been implemented. If you implement sockets in here this will be sitting in the wrapper. If you make the changes here and publish it to the library if any new application comes that we can directly use. So we need to be like uh careful which changes need to be done where. Okay. So actually entire UI authentication sockets everything is in Pog library. Whatever the UI changes are coming everything we are doing in the Pog Pog library itself. The major uh Salesforce prompt or uh tools we are writing it in here a enterprise plan. So in future if someone wants to integrate any other thing uh he can only write the single file and use pyog all features directly it will app UI will be written in pog ui all the UI will be written in pyog agent things should be written in this yes and is not in react or something it is completely in like uh HTML There are files all CSS files, HTML, CSS files are there. The plot is generating some iframe file and directly rendering in the UI like this. These all in iframe. This is just an iframe. This is all an iframe likewise. Okay. First try to set up it in your local then we'll see you guys might get questions I Rahul Mishra: Uh, Rajashekar G: think while setting it up. Rahul Mishra: yep. Rajashekar G: Okay, I think you don't everyone don't have the access to pilog also. Rahul Mishra: Pilog is open source right that yes Rajashekar G: No, Rahul Mishra: develop. Rajashekar G: he can't close. Transcription ended after 00:16:51 This editable transcript was computer generated and might contain errors. People can also change the text after it was created.