Mar 3, 2026
Enterprise Brain - V2 - Arch - Transcript
00:00:00
 
Yeshwanth Reddy Yerraguntla: Hi. Hello.
Vara Kumar Jagarapu: I
Yeshwanth Reddy Yerraguntla: Hi.
Vara Kumar Jagarapu: assume.
Yeshwanth Reddy Yerraguntla: Morning. Okay, I had a couple of demos enterprise brain wave then uh 10:00
Manisha Gundapuneedi: And now 5 minutes
Yeshwanth Reddy Yerraguntla: color
Manisha Gundapuneedi: document. Sorry.
Yeshwanth Reddy Yerraguntla: a document. No, no. Hey,
Manisha Gundapuneedi: Uh
Yeshwanth Reddy Yerraguntla: M
Manisha Gundapuneedi: connector
Yeshwanth Reddy Yerraguntla: email better. Mhm. Oh, miss notifications. You can't uh make Gmail into a REST API connector because Gmail contains embeddings. Gmail is all about the post but uh it is missing all the other components Right.
Manisha Gundapuneedi: like ID only just
Yeshwanth Reddy Yerraguntla: Screen share.
Manisha Gundapuneedi: connector document research exist.
Yeshwanth Reddy Yerraguntla: I saw that section SDK connector.
Manisha Gundapuneedi: SDK is nothing but Salesforce is Python package.
Yeshwanth Reddy Yerraguntla: End goal entity. Once we start this activity, what are we going to um say that we have accomplished?
 
 
00:07:04
 
Manisha Gundapuneedi: We have accomplished unified connector model. So where you will just give your uh config based on your connector and we should be able to pull the data. This is special.
Yeshwanth Reddy Yerraguntla: See, I understand the plan. In that case there's no point in sitting and generating these connectors when first of all there is no data connector. What I'm trying to say is uh why will we write a file connector if we don't have data to test it. So
Manisha Gundapuneedi: Last
Yeshwanth Reddy Yerraguntla: when the time comes. Anyway, if you have this in your mind, you can always extend on the existing uh connector.
Manisha Gundapuneedi: Rest two weeks.
Yeshwanth Reddy Yerraguntla: one second allow what I'm trying to say is yeah we will probably sit and create SDK connector but for which uh uh uh data source is my question I don't want to create a connector that has nothing to test against
Manisha Gundapuneedi: Salesforce
Yeshwanth Reddy Yerraguntla: Salesforce.
Manisha Gundapuneedi: already
Yeshwanth Reddy Yerraguntla: I can understand Gmail.
Manisha Gundapuneedi: and database Oracle group.
 
 
00:09:14
 
Yeshwanth Reddy Yerraguntla: Mhm. Because Nar was saying uh
Manisha Gundapuneedi: Enterprise Unifi client data
Yeshwanth Reddy Yerraguntla: something
Manisha Gundapuneedi: Oracle along with me are
Yeshwanth Reddy Yerraguntla: right.
Manisha Gundapuneedi: right.
Yeshwanth Reddy Yerraguntla: So what I what I will recommend is this is not a separate activity in my head from what I understood. Yeah. While you are creating this data source agent, the way you'll organize your code will fall into this format and mean there is no additional special
Manisha Gundapuneedi: Oh,
Yeshwanth Reddy Yerraguntla: status this needs to be given. What I'm trying to say is uh replace so let's say I'm in econom branch
Vara Kumar Jagarapu: Manisha, can you also share the documentar?
Yeshwanth Reddy Yerraguntla: uh cheetah. Yeah. Oh.
Vara Kumar Jagarapu: Sorry.
Yeshwanth Reddy Yerraguntla: Oh, sorry. Just
Vara Kumar Jagarapu: It's okay.
Yeshwanth Reddy Yerraguntla: got
Manisha Gundapuneedi: Ah, now better.
Yeshwanth Reddy Yerraguntla: what is happening? U anyway we have to create a bunch of connectors these data source agents current architecture the way it has been created is very suboptal right so if I go to backend uh database doc just a moment Oh, empty.
 
 
00:11:52
 
Yeshwanth Reddy Yerraguntla: Enterprise back app I'm sorry applo agents executing agent or first of all organization different it's not correct
Manisha Gundapuneedi: And I think maintenance syncing mechanism along with database maintenance for each connection.
Yeshwanth Reddy Yerraguntla: Yeah. When they are fetching connection for Salesforce hard right connect to insurance db config. Yeah this is coming from env. But once that is done uh uh
Manisha Gundapuneedi: connectionable define this function.
Yeshwanth Reddy Yerraguntla: variables actually we have to add this additional tool. Uh again we have to update the system prompt.
Manisha Gundapuneedi: Wait.
Yeshwanth Reddy Yerraguntla: Same thing with Salesforce. First of all we are getting again Salesforce table schema. Then we are running Salesforce query. Then there is a run insurance DB query. So we are duplicating our efforts again this is the pattern that is being proposed that let connectors have their own uh they they will whatever
Manisha Gundapuneedi: Oh,
Yeshwanth Reddy Yerraguntla: child classes parent class so that they will inherit from these guys so that our work is as minimal as possible.
 
 
00:13:44
 
Yeshwanth Reddy Yerraguntla: Okay, when you are tackling something, write that parent immediately and then uh write the child also so that
Manisha Gundapuneedi: heat.
Yeshwanth Reddy Yerraguntla: tomorrow if a new uh SDK based agent comes you have the parent uh to continue with as opposed to writing parents for every one of these four right today so that we are prepared. tested on we are only wasting our time and we don't know how efficient is our uh code
Manisha Gundapuneedi: requirement.
Yeshwanth Reddy Yerraguntla: in that case I'm saying this is a very small part of the whole this pipeline what you're writing is covering only these two guys
Manisha Gundapuneedi: Oh no.
Yeshwanth Reddy Yerraguntla: maximum I want to cover everything. So even if you want to start with SDK and API based connectors, I will not accept until you have everything. That's my point
Manisha Gundapuneedi: Okay. Connect
Yeshwanth Reddy Yerraguntla: that we'll have all the action items. So yesterday when I uh whatever time I spent 1 hour I came up with these as the MVP action items for Ganesh and Anita because they are the people who are working on uh UI
 
 
00:15:24
 
Manisha Gundapuneedi: Excuse
Yeshwanth Reddy Yerraguntla: part. Okati I have I want to discuss this with V and uh see if we can do something
Manisha Gundapuneedi: me.
Yeshwanth Reddy Yerraguntla: better or is this a decent starting point? Same with Manisha. Do you agree with this or not is what I want to discuss.
Manisha Gundapuneedi: Sure.
Yeshwanth Reddy Yerraguntla: First thing, second thing is you have to uh write the same list for other two modules also already. But now we are in the middle of the week.
Manisha Gundapuneedi: Okay.
Yeshwanth Reddy Yerraguntla: For me the highest importance is to freeze this guy and uh then we'll come to this point. This guy render Manisha's team will be u unblocked.
Manisha Gundapuneedi: Okay.
Yeshwanth Reddy Yerraguntla: Yeah. Okay, I will suggest uh uh for 5 10 minutes ask me questions. Until then I will try to write the list for uh data agents. Okay. discuss. Okay. Five 10 minutes while I come up with this.
 
 
00:17:13
 
Rajashekar G: And I'm trying to write the list for core agent.
Yeshwanth Reddy Yerraguntla: Oh no. Oh no.
Rajashekar G: Yeah.
Yeshwanth Reddy Yerraguntla: Anonymous. Now how can you edit these things? You have to create a whole copy. answer. Copy the whole thing into your own this thing.
Rajashekar G: Yes. Yes,
Yeshwanth Reddy Yerraguntla: Edit it.
Rajashekar G: I'll do
Yeshwanth Reddy Yerraguntla: Then you can paste it
Rajashekar G: it.
Yeshwanth Reddy Yerraguntla: back.
Rajashekar G: Okay. That's Hey, Hey, hey,
Vara Kumar Jagarapu: Uh user
Yeshwanth Reddy Yerraguntla: Yeah.
Vara Kumar Jagarapu: interfaces
Yeshwanth Reddy Yerraguntla: Huh? Oh,
Vara Kumar Jagarapu: how it is going to fit in here.
Yeshwanth Reddy Yerraguntla: no. One second. Uh
Vara Kumar Jagarapu: For example, user information right.
Yeshwanth Reddy Yerraguntla: oh, I can see.
Vara Kumar Jagarapu: So user
Yeshwanth Reddy Yerraguntla: Yeah.
Vara Kumar Jagarapu: interface
Yeshwanth Reddy Yerraguntla: User interfaces. Uh
Vara Kumar Jagarapu: something he asked but whether he is allowed or
 
 
00:29:43
 
Yeshwanth Reddy Yerraguntla: yeah.
Vara Kumar Jagarapu: not
Yeshwanth Reddy Yerraguntla: Oh no.
Vara Kumar Jagarapu: before the gray visualization.
Yeshwanth Reddy Yerraguntla: Correct. Correct. Ideally dashboard. Uh person corresponding
Rajashekar G: every data
Yeshwanth Reddy Yerraguntla: dashboard.
Vara Kumar Jagarapu: loop.
Rajashekar G: source.
Vara Kumar Jagarapu: Before coming to dashboard, we need to get store it somewhere.
Yeshwanth Reddy Yerraguntla: Uh
Vara Kumar Jagarapu: Next and login.
Yeshwanth Reddy Yerraguntla: yeah.
Vara Kumar Jagarapu: PC observated Uh key.
Yeshwanth Reddy Yerraguntla: observability.
Vara Kumar Jagarapu: Okay. Just
Yeshwanth Reddy Yerraguntla: Every one of these dubbers um needs to have certain inheritance properties.
Vara Kumar Jagarapu: Okay.
Yeshwanth Reddy Yerraguntla: All right. So uh side lo just intelligence architecture point is it needs to have all those uh items attached to it by
Vara Kumar Jagarapu: observally
Yeshwanth Reddy Yerraguntla: default. H. Yeah,
Vara Kumar Jagarapu: token. for
Yeshwanth Reddy Yerraguntla: correct. Got it.
Vara Kumar Jagarapu: uh whatever the functions
Yeshwanth Reddy Yerraguntla: Yes,
Vara Kumar Jagarapu: like MCP, right? So
 
 
00:32:26
 
Yeshwanth Reddy Yerraguntla: correct.
Vara Kumar Jagarapu: Okay. For example,
Yeshwanth Reddy Yerraguntla: system I want to show you just as an
Vara Kumar Jagarapu: okay.
Yeshwanth Reddy Yerraguntla: example last 30 days. So when an agent ran what was the trigger
Vara Kumar Jagarapu: Okay. Uh,
Yeshwanth Reddy Yerraguntla: question?
Vara Kumar Jagarapu: just give me a second.
Yeshwanth Reddy Yerraguntla: Oh, screen. I'm sorry.
Vara Kumar Jagarapu: Screen
Yeshwanth Reddy Yerraguntla: Can you see?
Vara Kumar Jagarapu: actually multiple screens.
Yeshwanth Reddy Yerraguntla: Multiple.
Vara Kumar Jagarapu: Okay. Uhhuh.
Yeshwanth Reddy Yerraguntla: Oh,
Vara Kumar Jagarapu: Not good.
Yeshwanth Reddy Yerraguntla: yeah. Okay.
Vara Kumar Jagarapu: Yeah. Yeah.
Yeshwanth Reddy Yerraguntla: Right.
Vara Kumar Jagarapu: No.
Yeshwanth Reddy Yerraguntla: So um for every agent what is the um input? What are the actually format? What is the input? This is the input. This is the next user asked this. Then assistant responded with this tool call.
Vara Kumar Jagarapu: Yeah.
Yeshwanth Reddy Yerraguntla: Then tool itself gave this as the output.
 
 
00:34:09
 
Vara Kumar Jagarapu: Okay.
Yeshwanth Reddy Yerraguntla: Then assistant took that input and uh created this output.
Vara Kumar Jagarapu: This is query query logging on everything
Yeshwanth Reddy Yerraguntla: Everything we are tracing on
Vara Kumar Jagarapu: right
Yeshwanth Reddy Yerraguntla: the uh the whole conversation we are tracing.
Vara Kumar Jagarapu: okay
Yeshwanth Reddy Yerraguntla: On top of that, we know how much money we are spending. How many input tokens of the tokens are we spending?
Vara Kumar Jagarapu: okay okay so right
Yeshwanth Reddy Yerraguntla: Yeah,
Vara Kumar Jagarapu: almosties
Yeshwanth Reddy Yerraguntla: correct.
Vara Kumar Jagarapu: Okay.
Yeshwanth Reddy Yerraguntla: Correct.
Vara Kumar Jagarapu: For example, particular skill point.
Yeshwanth Reddy Yerraguntla: particular skill point of view. Yeah.
Vara Kumar Jagarapu: Okay.
Yeshwanth Reddy Yerraguntla: Yeah. H this is the this is the red boxes which uh LLM is
Vara Kumar Jagarapu: Okay.
Yeshwanth Reddy Yerraguntla: getting uh uh triggered.
Vara Kumar Jagarapu: From the tool.
Yeshwanth Reddy Yerraguntla: Um so the which LLM is getting triggered running that is basically like which skill
Vara Kumar Jagarapu: Uh, okay.
Yeshwanth Reddy Yerraguntla: got triggered?
 
 
00:35:38
 
Vara Kumar Jagarapu: Okay. Okay. Nice. Then that part then that is
Yeshwanth Reddy Yerraguntla: uh uh what is the input sent to the
Vara Kumar Jagarapu: fine.
Yeshwanth Reddy Yerraguntla: agent? This is the whole Python code that agent decided to
Vara Kumar Jagarapu: Okay. So on top of
Yeshwanth Reddy Yerraguntla: run.
Vara Kumar Jagarapu: it, right?
Yeshwanth Reddy Yerraguntla: Correct.
Vara Kumar Jagarapu: Okay.
Yeshwanth Reddy Yerraguntla: Correct. Um
Vara Kumar Jagarapu: Okay. Stories related related
Yeshwanth Reddy Yerraguntla: yeah.
Vara Kumar Jagarapu: steps.
Yeshwanth Reddy Yerraguntla: So agent those few things are always constant when if you're bringing one agent which it means we are bringing other things also. Yeah. Next
Vara Kumar Jagarapu: training lesson. That is a kind of prompt that we are going to append to the query logologist
Yeshwanth Reddy Yerraguntla: Training lessons are though training
Vara Kumar Jagarapu: right
Yeshwanth Reddy Yerraguntla: lessons are more like um
Vara Kumar Jagarapu: kind of system level policies.
Yeshwanth Reddy Yerraguntla: Those are static policies.
 
 
00:37:26
 
Vara Kumar Jagarapu: Mhm.
Yeshwanth Reddy Yerraguntla: So how it should behave? Training lesson on the other hand is like um let me show you get into interactive mode. If I uh if I take this as an example here, I might say from now on uh sorry from now onwards uh I do not uh prefer um pie charts when there are more than 10 entities uh and pie chart for less than 10 entities. chat. There is no action to be taken by the agent. This is not a data query. This is a training
Vara Kumar Jagarapu: Yeah, understood. This is a from user level right?
Yeshwanth Reddy Yerraguntla: lesson whichever person.
Vara Kumar Jagarapu: User level training instruction.
Yeshwanth Reddy Yerraguntla: Yeah. Yeah.
Vara Kumar Jagarapu: Uh okay
Yeshwanth Reddy Yerraguntla: They have to go and sit in the form of its DB preferences somewhere it
Vara Kumar Jagarapu: for that
Yeshwanth Reddy Yerraguntla: has to uh somewhere for that user depends.
Vara Kumar Jagarapu: user.
 
 
00:38:44
 
Yeshwanth Reddy Yerraguntla: I can teach it for on behalf of the entire organization also. This if I am an elevated person,
Vara Kumar Jagarapu: Okay.
Yeshwanth Reddy Yerraguntla: this might be applicable to everyone in the organization going forward. Turn it person to
Vara Kumar Jagarapu: Mhm. Okay.
Yeshwanth Reddy Yerraguntla: Clarity.
Vara Kumar Jagarapu: lowle person I mean user level
Yeshwanth Reddy Yerraguntla: Oh,
Vara Kumar Jagarapu: person
Yeshwanth Reddy Yerraguntla: which one will take more president's shoes to
Vara Kumar Jagarapu: but value.
Yeshwanth Reddy Yerraguntla: right?
Vara Kumar Jagarapu: Just a
Yeshwanth Reddy Yerraguntla: Okay.
Vara Kumar Jagarapu: thought.
Yeshwanth Reddy Yerraguntla: Conflicting lessons.
Vara Kumar Jagarapu: training system. Those are the training on top of it. So policy there is nothing to change but then on top of it. Yeah. I also need to think
Yeshwanth Reddy Yerraguntla: Mhm. Last line.
Vara Kumar Jagarapu: uh starting training center whoever using this brain. So okay system they can do it.
Yeshwanth Reddy Yerraguntla: Right. Yeah,
 
 
00:41:43
 
Vara Kumar Jagarapu: At least not allow. All
Yeshwanth Reddy Yerraguntla: correct.
Vara Kumar Jagarapu: the
Yeshwanth Reddy Yerraguntla: But we started with a different question actually uh user uh information how it is being propagated.
Vara Kumar Jagarapu: propagated
Yeshwanth Reddy Yerraguntla: uh for every agent uh it needs to know which user is uh uh talking to
Vara Kumar Jagarapu: Sorry.
Yeshwanth Reddy Yerraguntla: it. So the uh start question how will it how will the agent fetch it or not?
Vara Kumar Jagarapu: Yeah. Yeah. the
Yeshwanth Reddy Yerraguntla: uh I don't know
Vara Kumar Jagarapu: results.
Yeshwanth Reddy Yerraguntla: the right mechanism but um somehow first thing is agents any LLF agent uh uh I mean at least in the framework that we write context uh it's possible to send
Vara Kumar Jagarapu: Hey
Yeshwanth Reddy Yerraguntla: force execution log sponder. Yeah, we have to take a slightly deeper look. But when some tool is getting run this agent first argument is always the context like which environment it is running who is running some environment information can be injected it's a it's basically a dictionary dictionary username other
 
 
00:43:44
 
Vara Kumar Jagarapu: Okay.
Yeshwanth Reddy Yerraguntla: things we can always like the moment you create this
Vara Kumar Jagarapu: Okay.
Yeshwanth Reddy Yerraguntla: ctx object llm will immediately know who it is dealing with and uh based on that uh the tools behavior will change. Uh so we can always customize the behavior. Only question is how will we send that signal? Exactly. But it's not impossible.
Rajashekar G: governance point we can also need to put it data source level also for so while getting the data from data
Yeshwanth Reddy Yerraguntla: Got it.
Rajashekar G: source itself it should fetch the right data for him like whatever he has access to
Yeshwanth Reddy Yerraguntla: Yeah.
Rajashekar G: salesforce so every data source
Yeshwanth Reddy Yerraguntla: Cool.
Rajashekar G: will be having a different data source Governance layers.
Yeshwanth Reddy Yerraguntla: Yeah. Basically what we need to send is uh uh we need to along with this also send u some additional context of who is sending this request on whose behalf is this request All right. I don't know.
 
 
00:45:35
 
Rajashekar G: And then user context
Yeshwanth Reddy Yerraguntla: I want it. I think login anybody can modify it. Yeah,
Rajashekar G: modify
Yeshwanth Reddy Yerraguntla: I think I gave everyone permission to write.
Rajashekar G: table. Clear complete.
Yeshwanth Reddy Yerraguntla: Great. Okay. Um Visualization life cycle.
Rajashekar G: My point is what needs to be visualiz agent has to give it to the UI
Yeshwanth Reddy Yerraguntla: Visualization intelligence. UI interface block number
Rajashekar G: interface block.
Yeshwanth Reddy Yerraguntla: two.
Rajashekar G: So agent will be giving the communication to another agent.
Yeshwanth Reddy Yerraguntla: Yeah, let this is the information.
Rajashekar G: Ch.
Yeshwanth Reddy Yerraguntla: So in the core agent there is block number two. Block number two will decide for every user what is the information it has to broadcast. So that is all it is sending for user uh ABC.
Rajashekar G: What do we got?
Yeshwanth Reddy Yerraguntla: These are the five items uh you have to put on UI the information pump to visualization agent.
 
 
00:49:05
 
Yeshwanth Reddy Yerraguntla: Visualization agent will consume those five items and figure out should it show a bar chart, should it show something
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: else.
Rajashekar G: So clearly
Yeshwanth Reddy Yerraguntla: box.
Rajashekar G: You can break
Yeshwanth Reddy Yerraguntla: H
Rajashekar G: each point.
Yeshwanth Reddy Yerraguntla: knowledge graph sync life cycle. Yeah. Shut up.
Vara Kumar Jagarapu: question.
Yeshwanth Reddy Yerraguntla: H. Oh no.
Vara Kumar Jagarapu: So,
Yeshwanth Reddy Yerraguntla: Okay. Basic chat. most basically better than Python second. Basically uh currently the way we handle it is we store all conversations in one big database against every user and all conversations are stored in the form of I think JSON like okay single JSON is one full conversation
Vara Kumar Jagarapu: user perspective like we can
Yeshwanth Reddy Yerraguntla: they're going Now let's go back to chat their chat history and continue from where they
Vara Kumar Jagarapu: store.
Yeshwanth Reddy Yerraguntla: stopped that we have to
Vara Kumar Jagarapu: Okay. So that is
Yeshwanth Reddy Yerraguntla: provide
Vara Kumar Jagarapu: Whatever it may be, dashboard, chart interface, user preferences,
 
 
00:59:42
 
Yeshwanth Reddy Yerraguntla: Mhm.
Vara Kumar Jagarapu: I user preferences and JP we are communicating to cor to get the reply, right?
Yeshwanth Reddy Yerraguntla: H
Vara Kumar Jagarapu: Uh, so request
Yeshwanth Reddy Yerraguntla: my user preferences. I will be able to tweak
Vara Kumar Jagarapu: for
Yeshwanth Reddy Yerraguntla: uh right. Okay, we can h
Vara Kumar Jagarapu: us.
Yeshwanth Reddy Yerraguntla: uh are there. So when I mean first thing is uh add these two things right training lessons and add our queries. Um we have to first decide on what kind of packet will be what kind of packet we will send we will send to the agents. So request headers if we can maintain which user is
Vara Kumar Jagarapu: Okay.
Yeshwanth Reddy Yerraguntla: sending then
Vara Kumar Jagarapu: For example,
Yeshwanth Reddy Yerraguntla: yeah uh
Vara Kumar Jagarapu: like based on his role, something needs to be if I something that needs to be based on I am
Yeshwanth Reddy Yerraguntla: Yeah.
Vara Kumar Jagarapu: what I am. So initial request.
Yeshwanth Reddy Yerraguntla: H.
Vara Kumar Jagarapu: Okay.
 
 
01:01:31
 
Yeshwanth Reddy Yerraguntla: Do you want to show it on screen like?
Vara Kumar Jagarapu: Uh just give me a second.
Yeshwanth Reddy Yerraguntla: Yeah. Rash MVP. So far it's looking good only.
Rajashekar G: Can you change
Yeshwanth Reddy Yerraguntla: Yeah. Yeah. Isolated
Rajashekar G: yourself?
Yeshwanth Reddy Yerraguntla: on
Vara Kumar Jagarapu: Just check.
Yeshwanth Reddy Yerraguntla: I need to miss myself.
Vara Kumar Jagarapu: Now access.
Yeshwanth Reddy Yerraguntla: As long as you log in, you can
Vara Kumar Jagarapu: Okay.
Yeshwanth Reddy Yerraguntla: modify.
Vara Kumar Jagarapu: So for example one case
Rajashekar G: All
Vara Kumar Jagarapu: right information. So it will give a
Yeshwanth Reddy Yerraguntla: Yeah,
Vara Kumar Jagarapu: response.
Rajashekar G: right.
Yeshwanth Reddy Yerraguntla: preferences login information.
Vara Kumar Jagarapu: login information. Okay. User
Yeshwanth Reddy Yerraguntla: Yeah.
Vara Kumar Jagarapu: username username.
Yeshwanth Reddy Yerraguntla: Right.
Vara Kumar Jagarapu: Okay. So access token
Yeshwanth Reddy Yerraguntla: H
Vara Kumar Jagarapu: something which is available.
Yeshwanth Reddy Yerraguntla: correct.
Vara Kumar Jagarapu: something. Okay.
 
 
01:04:53
 
Vara Kumar Jagarapu: So
Yeshwanth Reddy Yerraguntla: Correct.
Vara Kumar Jagarapu: checking
Yeshwanth Reddy Yerraguntla: Correct.
Vara Kumar Jagarapu: understanding.
Yeshwanth Reddy Yerraguntla: Got it.
Vara Kumar Jagarapu: Okay, thank
Yeshwanth Reddy Yerraguntla: Yeah.
Vara Kumar Jagarapu: you.
Rajashekar G: Right now we are using ha user info key direct key.
Yeshwanth Reddy Yerraguntla: Hi
Rajashekar G: Uh we are using for user
Yeshwanth Reddy Yerraguntla: Kuna
Rajashekar G: information
Yeshwanth Reddy Yerraguntla: as an example.
Rajashekar G: current somewhere in the preferences table we are mapping it with users.
Yeshwanth Reddy Yerraguntla: Correct.
Rajashekar G: Okay. Okay. I like it.
Vara Kumar Jagarapu: knowledge. Hello.
Yeshwanth Reddy Yerraguntla: mute lo I'm sorry. Yeah.
Vara Kumar Jagarapu: What exactly and how how it will be stored the information.
Yeshwanth Reddy Yerraguntla: Um but it's going to be a collection of nodes and it's a graph data structure. Nodes adj are talking
Vara Kumar Jagarapu: Okay. generally vector completely
Yeshwanth Reddy Yerraguntla: about
Vara Kumar Jagarapu: uh
Yeshwanth Reddy Yerraguntla: vector DV is different in the sense that uh every row has some factual
Vara Kumar Jagarapu: different
 
 
01:21:37
 
Yeshwanth Reddy Yerraguntla: information in vector DV and against that factual information you'll have some um embedding associated embedding is a way to do fast lookup. Uh so the question is like what are all the activities
Vara Kumar Jagarapu: H.
Yeshwanth Reddy Yerraguntla: in project infinitism vector store might give you all the 10 or 20 facts it might be sufficient for us it might not be I don't know but one level on top of embedding uh graph databases sorry rack databases graph Here against every row you also have the edge information how the rows are connected to each other.
Vara Kumar Jagarapu: Okay.
Yeshwanth Reddy Yerraguntla: So that will uh that might help us with uh uh connection connectivity based uh information. It's it's it's a possibility that it can do that.
Vara Kumar Jagarapu: H.
Yeshwanth Reddy Yerraguntla: But do we need graph or simple embedding uh rag DV sufficient uh from at least from
Vara Kumar Jagarapu: Okay.
Yeshwanth Reddy Yerraguntla: an MVP perspective? It is probably sufficient to create uh just a normal rag DB.
Vara Kumar Jagarapu: Okay. So I think the more information we have related information vector D is more useful.
 
 
01:23:29
 
Vara Kumar Jagarapu: Okay.
Yeshwanth Reddy Yerraguntla: Yes.
Rajashekar G: is working on
Yeshwanth Reddy Yerraguntla: Now let's craft
Rajashekar G: Guit.
Yeshwanth Reddy Yerraguntla: for stories.
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: On top of that, he said uh sorry data
Rajashekar G: data
Yeshwanth Reddy Yerraguntla: encryption.
Rajashekar G: encryption.
Yeshwanth Reddy Yerraguntla: Oh no exactly
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: connect. He said then I said deliver it and come back to architecture discussion. So we
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: do
Rajashekar G: Okay. And so like
Yeshwanth Reddy Yerraguntla: to launch.
Rajashekar G: But they will be working on like a enterprise brand. Okay.
Yeshwanth Reddy Yerraguntla: Yeah.
Rajashekar G: Enterprise brands whatever we have done.
Yeshwanth Reddy Yerraguntla: Uh
Rajashekar G: Okay. Okay. Next. I should
Yeshwanth Reddy Yerraguntla: Mhm.
Rajashekar G: have flickering.
Yeshwanth Reddy Yerraguntla: I like
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: every
Rajashekar G: Too much
Yeshwanth Reddy Yerraguntla: but after 5 seconds uh it's okay. Nobody cares the amount of effort we are putting on that it's waste.
 
 
01:28:14
 
Rajashekar G: next react.
Yeshwanth Reddy Yerraguntla: I don't
Rajashekar G: So
Yeshwanth Reddy Yerraguntla: know.
Rajashekar G: I should say current system that's a global history user specific It is.
Yeshwanth Reddy Yerraguntla: current system current
Rajashekar G: Yes.
Yeshwanth Reddy Yerraguntla: user specific. Yeah, user specific on Oh, you're saying everybody can access everyone's conversations. Uh yeah,
Rajashekar G: Yes.
Yeshwanth Reddy Yerraguntla: important broccult.
Rajashekar G: Someone has to give something.
Yeshwanth Reddy Yerraguntla: Someone can access everybody's conversations.
Rajashekar G: Oh, has been changed now.
Yeshwanth Reddy Yerraguntla: Yeah. Yeah. change.
Rajashekar G: Okay. And uh that should be into pilot
Yeshwanth Reddy Yerraguntla: Huh?
Rajashekar G: color.
Yeshwanth Reddy Yerraguntla: Uh okay. Who? Whichever user is uh logging in while associated.
Rajashekar G: No.
Yeshwanth Reddy Yerraguntla: Yeah. Yeah.
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: You ever Oh,
Rajashekar G: Choose point.
Yeshwanth Reddy Yerraguntla: heat.
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: And
Rajashekar G: Running Python. U fetching information.
Yeshwanth Reddy Yerraguntla: yeah
 
 
01:30:55
 
Rajashekar G: Generating the modification
Yeshwanth Reddy Yerraguntla: purpose keyword adj. Yes. Against every
Rajashekar G: change.
Yeshwanth Reddy Yerraguntla: Yeah.
Rajashekar G: next to. Okay.
Yeshwanth Reddy Yerraguntla: Does this work?
Rajashekar G: morning.
Yeshwanth Reddy Yerraguntla: I don't
Rajashekar G: So,
Yeshwanth Reddy Yerraguntla: know. Uh, got it.
Rajashekar G: While deployment local it's working but while deployment breaking. So initially it was public told me he made it
Yeshwanth Reddy Yerraguntla: Um,
Rajashekar G: private. So make it public. So to bypass this either we need to use
Yeshwanth Reddy Yerraguntla: okay.
Rajashekar G: public
Yeshwanth Reddy Yerraguntla: I mean it's just a UI agent so I don't know it's not like there's anything dewarming specific in it. Um, cut them.
Rajashekar G: I'm done
Yeshwanth Reddy Yerraguntla: Yeah.
Rajashekar G: manufacturing agent.py Salesforce.py by here is
Yeshwanth Reddy Yerraguntla: the connector.
Rajashekar G: okay.
Yeshwanth Reddy Yerraguntla: He
Rajashekar G: Next
Yeshwanth Reddy Yerraguntla: Yeah, every day on
Rajashekar G: demo.
Yeshwanth Reddy Yerraguntla: Friday.
Rajashekar G: Hi
Satyasri Prabhakar Mantripragada: Hey,
Rajashekar G: sir.
 
 
01:34:22
 
Satyasri Prabhakar Mantripragada: good morning.
Yeshwanth Reddy Yerraguntla: Don't burn
Satyasri Prabhakar Mantripragada: Good. Requirement.
Yeshwanth Reddy Yerraguntla: pricing. Whoever is going through it, they are liking it for sure.
Satyasri Prabhakar Mantripragada: Okay.
Yeshwanth Reddy Yerraguntla: uh two clients presentation and a second client was asking what is the pricing like how long will you able to deliver it in all questions
Satyasri Prabhakar Mantripragada: Okay. Do we get a chance to ask for this test data or some kind of production data?
Yeshwanth Reddy Yerraguntla: sorry
Satyasri Prabhakar Mantripragada: Did we get a chance to ask the production data or some kind of test data?
Yeshwanth Reddy Yerraguntla: yeah I at that time I think they will like sorry morning. Yeah,
Satyasri Prabhakar Mantripragada: I think.
Yeshwanth Reddy Yerraguntla: the action items have loop.
Satyasri Prabhakar Mantripragada: So I think Raj for visual effect detection right if we are connecting with any client and if we can get their test data the expectation is that we'll simulate that within we connect that within ouration and then get their insight for their data
Yeshwanth Reddy Yerraguntla: True.
 
 
01:36:06
 
Rajashekar G: Oh,
Yeshwanth Reddy Yerraguntla: True.
Satyasri Prabhakar Mantripragada: like
Rajashekar G: all
Yeshwanth Reddy Yerraguntla: We should be in that
Rajashekar G: right.
Yeshwanth Reddy Yerraguntla: position.
Satyasri Prabhakar Mantripragada: Quick
Rajashekar G: Yeah.
Satyasri Prabhakar Mantripragada: turn.
Yeshwanth Reddy Yerraguntla: Yeah.
Rajashekar G: It takes doesn't take much time. technically simply and is it lesser
Yeshwanth Reddy Yerraguntla: Yeah.
Satyasri Prabhakar Mantripragada: Okay.
Rajashekar G: Okay.
Satyasri Prabhakar Mantripragada: Right.
Yeshwanth Reddy Yerraguntla: Yeah. At least to understand the data. 4 hours should be enough to come up with the story.
Satyasri Prabhakar Mantripragada: Okay,
Yeshwanth Reddy Yerraguntla: It might uh definitely take more than 8 to 12 hours because there might be lot of
Rajashekar G: Sucks.
Yeshwanth Reddy Yerraguntla: slices uh lot of experimenting that you have to do.
Satyasri Prabhakar Mantripragada: so yeah, the reason that I was asking is if we take less than 1 uh 2 hours that the system is intelligent enough to craft a story quickly then I thought let us wait a bit in the night so that the moment they give it we'll take 1 hour 2 hours time and then we so that shows the acceleration but if it take more 4 hours and there's no
 
 
01:38:05
 
Yeshwanth Reddy Yerraguntla: I see solve it. We at least once we can try that one to our activity. data set I think is a waste of time.
Satyasri Prabhakar Mantripragada: Okay.
Yeshwanth Reddy Yerraguntla: So when they give it let's try to sit for one two hours and see how much we can deliver but we'll not promise we'll we know how much we can do in limited time. So uh
Satyasri Prabhakar Mantripragada: And accordingly we can set the expectations with management.
Yeshwanth Reddy Yerraguntla: pra at least once we once we do it ourselves then we can set the expectations
Satyasri Prabhakar Mantripragada: Yeah.
Yeshwanth Reddy Yerraguntla: on
Satyasri Prabhakar Mantripragada: Yeah.
Yeshwanth Reddy Yerraguntla: we are sort of that like at least uh to come up with an MVP for the junior Right. Even though we are writing the components, what we feel is either it's too vague or too broad.
Satyasri Prabhakar Mantripragada: Okay.
Yeshwanth Reddy Yerraguntla: So,
Satyasri Prabhakar Mantripragada: Okay. Uh can you give an example?
Yeshwanth Reddy Yerraguntla: uh example I can give is um um so first thing is for the three major blocks we have right UI block okay core agent block data source block we have isolated um what are the life cycles what are the components and uh Jaggle game for each subsystem we try to create a single MVP.
 
 
01:40:30
 
Yeshwanth Reddy Yerraguntla: Sometimes um as you can see right if the life
Satyasri Prabhakar Mantripragada: Okay.
Yeshwanth Reddy Yerraguntla: cycle is about context propagation where a user is asking questions and getting back answers. MVP in two weeks I wrote a thing called create a standalone react chat app with copilot kit interface that supports widgets. How much will they able to understand? Obviously, they'll be confused.
Satyasri Prabhakar Mantripragada: Okay.
Yeshwanth Reddy Yerraguntla: Strong feeling that we have to go one level deeper and Ganesh or Abila Sham for the next two
Satyasri Prabhakar Mantripragada: Okay. Oh god.
Yeshwanth Reddy Yerraguntla: weeks. Similarly, we take some other component like uh alert life cycle only send email alert when threshold crossed basic acknowledgement tracking and even but this is so vague that definitely they'll only come back to us and say especially when we are trying to handle 10 people sorry eight
Satyasri Prabhakar Mantripragada: Okay. So and then um and then for the alert right for example uh how
Yeshwanth Reddy Yerraguntla: open.
Satyasri Prabhakar Mantripragada: I would approach you simulate an alert uh and then
 
 
01:41:52
 
Yeshwanth Reddy Yerraguntla: Yeah.
Satyasri Prabhakar Mantripragada: you monitoring something and triggering an alert phaser phase or The entire thing is cut into multiple steps. The first step is monitoring some uh entity and on a certain time it
Yeshwanth Reddy Yerraguntla: Yeah.
Satyasri Prabhakar Mantripragada: would trigger an alert that something went wrong simulation. Second system you'll develop is monitoring that alert that was generated by somebody else and then
Yeshwanth Reddy Yerraguntla: H.
Satyasri Prabhakar Mantripragada: shooting an email. Third one is how do you craft an email based on the event that was triggered the alert that was triggered
Yeshwanth Reddy Yerraguntla: Okay. Okay.
Satyasri Prabhakar Mantripragada: alert number of weert type of severity. So at least it does not matter what kind of alert comes out.
Yeshwanth Reddy Yerraguntla: Mhm.
Satyasri Prabhakar Mantripragada: But what you have to do is given an alert your your job is just to initiate an email. So you can break that into maybe my thought
Yeshwanth Reddy Yerraguntla: are the understood just level. So I was going too far in trying to come up with the specification itself but you're
 
 
01:43:32
 
Satyasri Prabhakar Mantripragada: process.
Yeshwanth Reddy Yerraguntla: saying just break that one thing down into multiple steps and uh uh so that at least those steps are somewhat easy to digest.
Satyasri Prabhakar Mantripragada: Don't
Yeshwanth Reddy Yerraguntla: Break out.
Satyasri Prabhakar Mantripragada: put
Rajashekar G: They go to
Yeshwanth Reddy Yerraguntla: Now
Manisha Gundapuneedi: Nothing.
Rajashekar G: them.
Satyasri Prabhakar Mantripragada: Oh,
Vara Kumar Jagarapu: I'm not it.
Satyasri Prabhakar Mantripragada: this
Rajashekar G: Hold
Yeshwanth Reddy Yerraguntla: Okay. Ah.
Rajashekar G: them.
Yeshwanth Reddy Yerraguntla: Hello.
Satyasri Prabhakar Mantripragada: is
Yeshwanth Reddy Yerraguntla: s I don't know connection we'll wait for 5 minutes I interest on the break. Now I want to hear from Manisha Rashakar and how to give it to these people so that uh it's not too vague or it's not too chances of success. As simple as
Vara Kumar Jagarapu: Are there anything?
Manisha Gundapuneedi: And they still on the same lines.
Yeshwanth Reddy Yerraguntla: that.
Manisha Gundapuneedi: Hello. Hello.
Rajashekar G: Elaborate chase.
Vara Kumar Jagarapu: elaborate.
Rajashekar G: We will have a complete discussion with all the team
 
 
01:46:08
 
Vara Kumar Jagarapu: So at
Rajashekar G: alone.
Vara Kumar Jagarapu: least
Yeshwanth Reddy Yerraguntla: H.
Vara Kumar Jagarapu: they'll get more into it.
Yeshwanth Reddy Yerraguntla: Sir uh from all the three responses I got right from Praakar Suru,
Vara Kumar Jagarapu: Thank you.
Yeshwanth Reddy Yerraguntla: V and Manisha,
Vara Kumar Jagarapu: Thank
Yeshwanth Reddy Yerraguntla: what I understand is first thing once we agree to that this is
Vara Kumar Jagarapu: you.
Yeshwanth Reddy Yerraguntla: a decent MVP V like at least one V3 should be on the same page that this is decent and once that is done we can break it down into multiple uh sub uh uh instructions for every MAP there will be three to five instructions on step one step two step three step five at that point I think we'll have enough clarity to get into technical details also probably right.
Rajashekar G: uh I have one more point like
Yeshwanth Reddy Yerraguntla: H.
Rajashekar G: so whenever we are going to do this the first should be like first priority whatever tool comes that needs to be done in four weeks or something.
Yeshwanth Reddy Yerraguntla: All
 
 
01:47:52
 
Rajashekar G: So that
Yeshwanth Reddy Yerraguntla: right.
Rajashekar G: MVP what needs to be done. What needs to be done? 20 points.
Manisha Gundapuneedi: Oracle Unifi
Rajashekar G: Yes.
Manisha Gundapuneedi: the system should already have the knowledge like same it should already have the supply chain management knowledge and it should have the ability to learn.
Yeshwanth Reddy Yerraguntla: Okay.
Manisha Gundapuneedi: It should have a chart interface where user will be able to chart and get answers
Rajashekar G: All
Manisha Gundapuneedi: only. It should be able
Yeshwanth Reddy Yerraguntla: Okay.
Manisha Gundapuneedi: to
Rajashekar G: right.
Yeshwanth Reddy Yerraguntla: Okay.
Rajashekar G: Dynamic UI dynamic UI
Manisha Gundapuneedi: Some
Rajashekar G: training escalation email sending a three
Manisha Gundapuneedi: level
Rajashekar G: point.
Manisha Gundapuneedi: five points.
Yeshwanth Reddy Yerraguntla: Yeah. Yeah.
Rajashekar G: That should be like one of the first
Yeshwanth Reddy Yerraguntla: Okay then.
Rajashekar G: milestone.
Manisha Gundapuneedi: What
Yeshwanth Reddy Yerraguntla: Check.
Manisha Gundapuneedi: about
Yeshwanth Reddy Yerraguntla: Now explain what is that uh simulator PC
Manisha Gundapuneedi: so simulator
Yeshwanth Reddy Yerraguntla: doing.
Manisha Gundapuneedi: P discussion Hello framework framework
 
 
01:49:59
 
Yeshwanth Reddy Yerraguntla: Okay.
Manisha Gundapuneedi: based on the scenario and user that is logging in system should be able to generate dynamic dashboard. So compenario system based on the framework system should be able to say okay this is how your UI is going to
Yeshwanth Reddy Yerraguntla: Mhm.
Manisha Gundapuneedi: model. Last two days.
Yeshwanth Reddy Yerraguntla: Okay. Okay. Okay.
Manisha Gundapuneedi: with Gemini model.
Yeshwanth Reddy Yerraguntla: Got it.
Manisha Gundapuneedi: Gemini model
Yeshwanth Reddy Yerraguntla: Got it. Mhm.
Manisha Gundapuneedi: Snapshot of data
Yeshwanth Reddy Yerraguntla: See um I'm just highlighting a few things because um escalation life cycle kali because like these are the things that unifier sorry data still will be interested in right. So orange governance is like critical everything else the governance and designs are critical. Uh the other uh whatever blocks I highlighted in yellow. Uh just highlight just the 1 2 3 4 5 6 7 8 9.
Rajashekar G: Yes.
Yeshwanth Reddy Yerraguntla: probably we can just focus on these nine guys and uh um agree
 
 
01:52:35
 
Satyasri Prabhakar Mantripragada: Feel
Yeshwanth Reddy Yerraguntla: that these are worth doing in the next two weeks uh and give them to each person. So shall we go through them?
Rajashekar G: Yeah, sure.
Yeshwanth Reddy Yerraguntla: We can say that we we'll focus on these items.
Rajashekar G: It's
Satyasri Prabhakar Mantripragada: that.
Yeshwanth Reddy Yerraguntla: Yeah. Unifier in a way.
Satyasri Prabhakar Mantripragada: Oh,
Yeshwanth Reddy Yerraguntla: So let's just go through them. Um then we can decide uh how to structure the MVP.
Satyasri Prabhakar Mantripragada: okay.
Yeshwanth Reddy Yerraguntla: The first thing is uh the most basic functionality is ad hoc questioning and answering. Ad hoc is uh what we always do uh in according to our definition. This is this asking question and getting back answer is what I am calling as ad hoc. So the agents or components that it is touching is the chat interface. Chat interface sends message to core agent. Core agent whatever it is doing it has to do. I'm not saying core agent will directly send back to UI agent.
 
 
01:54:01
 
Yeshwanth Reddy Yerraguntla: Core agent will do some processing with data agents or its own knowledge base. It doesn't matter. But core agent will send back some information to UI agent. UI agent will put it back on chat. uh the information traveling across all the components is you have ad hoc query, you have ad hoc answers, you have visualizations, you can I can call it ad hoc visualizations, we have user context as well. anything we are missing and the MVP that I am proposing is create a standalone react chat app with co-pilot interface that supports widgets right um I mean this is where I currently
Rajashekar G: multiple points we need to list like chat
Yeshwanth Reddy Yerraguntla: Mhm.
Rajashekar G: app needs to be like streaming enabled sockets integrated like uh UI components UI component related and related like involves last time.
Yeshwanth Reddy Yerraguntla: Okay. Good.
Rajashekar G: So we need to have all the conversations with Ward also to finalize few things and architecture the front end
Yeshwanth Reddy Yerraguntla: Okay.
 
 
01:55:27
 
Yeshwanth Reddy Yerraguntla: And
Rajashekar G: completely.
Yeshwanth Reddy Yerraguntla: got it.
Rajashekar G: Authentication handling
Yeshwanth Reddy Yerraguntla: Front authentic. Yeah.
Rajashekar G: react authentication. So chart interface
Yeshwanth Reddy Yerraguntla: Okay.
Rajashekar G: content points. So,
Yeshwanth Reddy Yerraguntla: Uh so this is not MVP anymore, right? Whatever you're saying chat history
Rajashekar G: Contestation
Yeshwanth Reddy Yerraguntla: game
Rajashekar G: center context.
Yeshwanth Reddy Yerraguntla: to a large extent. Yeah, we just doing this in two weeks is a decent activity for them even though we writing other things. Uh what else? Uh streaming sockets uh visualization
Rajashekar G: I don't think you
Yeshwanth Reddy Yerraguntla: components MVP instructions and make notes. essentially whatever we have done in.
Rajashekar G: We can go
Yeshwanth Reddy Yerraguntla: No sorry
Rajashekar G: in point error handling front end error handling That's that should be inherent. But
Yeshwanth Reddy Yerraguntla: level. It just fails. Call info.
Rajashekar G: it's not anything to the
Yeshwanth Reddy Yerraguntla: Oh no.
Rajashekar G: user.
 
 
01:57:59
 
Yeshwanth Reddy Yerraguntla: Yeah. Sorry before we jump to other things and decide okay let's go through other things also like this
Satyasri Prabhakar Mantripragada: one minute all the things that we need to
Yeshwanth Reddy Yerraguntla: is
Satyasri Prabhakar Mantripragada: do we know what are all the things that we need
Yeshwanth Reddy Yerraguntla: Yeah.
Satyasri Prabhakar Mantripragada: to do I don't consider it as MVP or product but for an enterprise as a platform
Yeshwanth Reddy Yerraguntla: Yes.
Satyasri Prabhakar Mantripragada: what do we need to do once that is done let us also package them from the to
Yeshwanth Reddy Yerraguntla: Correct. Correct. Correct. We'll do
Satyasri Prabhakar Mantripragada: most.
Yeshwanth Reddy Yerraguntla: that. Yeah.
Satyasri Prabhakar Mantripragada: So next level is then let us map them to the resources based on their capability. This is how we
Yeshwanth Reddy Yerraguntla: Okay. What else are you saying?
Satyasri Prabhakar Mantripragada: make
Yeshwanth Reddy Yerraguntla: Priority, right? Priority.
Satyasri Prabhakar Mantripragada: into the same loop.
Yeshwanth Reddy Yerraguntla: This is what I was missing.
 
 
01:59:53
 
Yeshwanth Reddy Yerraguntla: Um yeah. Yeah. Yeah. Do that.
Satyasri Prabhakar Mantripragada: It is very clear for them.
Yeshwanth Reddy Yerraguntla: Yeah. True.
Rajashekar G: All
Yeshwanth Reddy Yerraguntla: Yeah. Let's continue. Uh context propagation life cycle basically question answering context who is asking the questions u
Rajashekar G: right.
Yeshwanth Reddy Yerraguntla: and message history. Technically these will come here. Hot basic uh chat
Rajashekar G: What you mean?
Yeshwanth Reddy Yerraguntla: interface just by itself we'll have a bunch of these things right
Rajashekar G: All
Yeshwanth Reddy Yerraguntla: streaming socket. So visualization components everything we have done
Rajashekar G: right.
Yeshwanth Reddy Yerraguntla: in pyog which means question and answer life cycle. What are the new things that uh we need to show a question answering?
Rajashekar G: UA agent will be talking with the core agent,
Yeshwanth Reddy Yerraguntla: Yeah.
Rajashekar G: right?
Yeshwanth Reddy Yerraguntla: Core agent talks to UI. UI doesn't talk to
Rajashekar G: All right.
Yeshwanth Reddy Yerraguntla: core.
 
 
02:01:52
 
Rajashekar G: Core agent will be giving the instructions to UA agent. What needs to be rendered on the UI or core agent is giving only the
Yeshwanth Reddy Yerraguntla: Correct. No.
Rajashekar G: answer.
Yeshwanth Reddy Yerraguntla: Yes,
Rajashekar G: U agent is deciding what to render.
Yeshwanth Reddy Yerraguntla: perfect.
Rajashekar G: set of instructions. What are all the visualation tools it has capabilities?
Yeshwanth Reddy Yerraguntla: Yes.
Rajashekar G: Input structure and output structure.
Yeshwanth Reddy Yerraguntla: even I I agree with what you just
Rajashekar G: So UA agent agentic
Yeshwanth Reddy Yerraguntla: said.
Rajashekar G: setup a pentic pyic UA agent setup I'm saying it like a life cycle on a sub parts of components
Vara Kumar Jagarapu: What are the components involved?
Rajashekar G: Then I'd say
Yeshwanth Reddy Yerraguntla: Add a question answer life cycle. What's the features? Got user preferences.
Rajashekar G: over the same level.
Yeshwanth Reddy Yerraguntla: I didn't type this by mistake.
Rajashekar G: Okay. System
Yeshwanth Reddy Yerraguntla: Mhm.
Rajashekar G: prompt.
Yeshwanth Reddy Yerraguntla: System I don't know agent that I have have six boxes, right?
 
 
02:04:21
 
Rajashekar G: Right.
Yeshwanth Reddy Yerraguntla: any agent you are about to write will it should automatically have all of these
Rajashekar G: All
Yeshwanth Reddy Yerraguntla: things right so you are not specifically saying I will write
Rajashekar G: right.
Yeshwanth Reddy Yerraguntla: instructions I will write these things nothing like that okay parent class calling we are calling it as intelligence or something that parent class should
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: automatically have all of these things uh built in right
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: and then inherit
Rajashekar G: So simple agent
Yeshwanth Reddy Yerraguntla: Exactly.
Rajashekar G: Ident.
Yeshwanth Reddy Yerraguntla: System
Rajashekar G: Got it. Skills are nothing but tools.
Yeshwanth Reddy Yerraguntla: skills are tools ra you'll have skills brackets tools tools
Rajashekar G: Uh particular where is system prompt or
Yeshwanth Reddy Yerraguntla: book Drag system
Rajashekar G: something?
Yeshwanth Reddy Yerraguntla: prompt I mean drag it's just a database of all these things cash training lessons all those things will come resources. training lessons on the other hand is like uh user persona Not.
 
 
02:07:35
 
Yeshwanth Reddy Yerraguntla: I think training cycle like all it needs to say is who is user person who is trainer person
Rajashekar G: Right.
Yeshwanth Reddy Yerraguntla: user preferences continue you writing something I render
Rajashekar G: I think
Yeshwanth Reddy Yerraguntla: Exactly. Oh
Rajashekar G: points widget.
Yeshwanth Reddy Yerraguntla: no.
Rajashekar G: So any agent we are willing like um if there
Yeshwanth Reddy Yerraguntla: Yeah,
Rajashekar G: questions us it should ask the follow
Yeshwanth Reddy Yerraguntla: follow questions.
Rajashekar G: question inbuilt
Yeshwanth Reddy Yerraguntla: Yeah, good point. Question will trigger, right?
Rajashekar G: followup question but this point needs to needs to be in the this block even whatever we are maintaining right six
Yeshwanth Reddy Yerraguntla: Facebook
Rajashekar G: blocks it's
Yeshwanth Reddy Yerraguntla: confusion
Rajashekar G: Uh, that's wrong actually.
Yeshwanth Reddy Yerraguntla: can only come from userus.
Rajashekar G: Right. Right.
Yeshwanth Reddy Yerraguntla: that user is the only person who can clarify in those
Rajashekar G: Oh no.
Yeshwanth Reddy Yerraguntla: cases.
Rajashekar G: Oh no.
Yeshwanth Reddy Yerraguntla: Um, so two more features I just recollected tool call info type not just in questions in followup as followup questions and so you can see my screen
 
 
02:09:55
 
Rajashekar G: Right.
Yeshwanth Reddy Yerraguntla: right action total
Rajashekar G: Right.
Yeshwanth Reddy Yerraguntla: estimated so blah blah blah Would you like me to do this?
Rajashekar G: Oh
Yeshwanth Reddy Yerraguntla: That can act as a pill.
Rajashekar G: no.
Yeshwanth Reddy Yerraguntla: These are the four followup nudges that will look good for user
Rajashekar G: Right.
Yeshwanth Reddy Yerraguntla: interaction.
Rajashekar G: Right.
Yeshwanth Reddy Yerraguntla: So now
Rajashekar G: Just as followup questions. Thumbs down.
Yeshwanth Reddy Yerraguntla: Right.
Rajashekar G: It's not even not yet push it to this enterprise brain yet.
Yeshwanth Reddy Yerraguntla: True. Yeah.
Rajashekar G: See that one also can be
Yeshwanth Reddy Yerraguntla: True.
Rajashekar G: there.
Yeshwanth Reddy Yerraguntla: So, thumbs up, thumbs down, kodism. APS to send mail and WhatsApp critical messages service to uh Some service should call this guy end of the day news and insights life cycle design I don't know design is mainly coming from CSS S for specs for visualizations UX dashboard widgets.
Rajashekar G: something inside. New showcase
Yeshwanth Reddy Yerraguntla: Mhm.
 
 
02:12:28
 
Rajashekar G: only,
Yeshwanth Reddy Yerraguntla: Yeah. See the UI agent is intercepting the news. So main the way I would look at it is uh receive news
Rajashekar G: Right.
Yeshwanth Reddy Yerraguntla: from core agent change and the UI uh UI should uh uh new ro agent for which user decide based on user pref Uh what to show in dashboard? Uh show it up in dashboard. New showcase UI is basically dashboard only. I mean are we missing anything additionally here? full decide skills. Okay.
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: Next like I think we are going at in the right
Rajashekar G: Sure.
Yeshwanth Reddy Yerraguntla: direction now agent.
Rajashekar G: And others may add points.
Yeshwanth Reddy Yerraguntla: Yeah, adding direct discussion group. If you think something is worth adding direct double click add. So
Rajashekar G: I think these are all we have in mind.
Yeshwanth Reddy Yerraguntla: yeah.
Rajashekar G: Come on.
Manisha Gundapuneedi: actually address.
Yeshwanth Reddy Yerraguntla: So context propagation life cycle what are we doing dashboard chat realtime intelligence session store update snapshot user snapshot snapshot user snapshot
 
 
02:15:39
 
Rajashekar G: I mean here context propagation life cycle the agent
Yeshwanth Reddy Yerraguntla: Raj.
Rajashekar G: should have the knowledge of all these and Answer.
Yeshwanth Reddy Yerraguntla: Uh, this answer correct.
Rajashekar G: Knowledge. Knowledge
Yeshwanth Reddy Yerraguntla: No,
Rajashekar G: base.
Yeshwanth Reddy Yerraguntla: no, no, no, no. One second. One second. Next.
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: Oh, user interacted something on dashboard. User has something in chat. Add a query real time in.
Rajashekar G: Oh no. What's it?
Yeshwanth Reddy Yerraguntla: Okay. Then you are going into sessions. Session story.
Rajashekar G: session story
Yeshwanth Reddy Yerraguntla: Okay.
Rajashekar G: previous
Yeshwanth Reddy Yerraguntla: Okay. User snapshot is uh user snapshot which user asked the question or snapshot or information.
Rajashekar G: kind
Yeshwanth Reddy Yerraguntla: Okay. User snapshot. It doesn't matter which user is asking. Uh okay. It does matter.
Rajashekar G: Maybe it matter like even if it is an
Yeshwanth Reddy Yerraguntla: Correct.
 
 
02:17:00
 
Rajashekar G: NLP to SQL real time intelligence agent should uh develop an
Yeshwanth Reddy Yerraguntla: Yeah.
Rajashekar G: SQL query that's it but as JS time it should know who is the user and what are
Yeshwanth Reddy Yerraguntla: Correct.
Rajashekar G: his preferences and organization a level Which state or it's like it's normal direct querying on the database and giving the answer.
Yeshwanth Reddy Yerraguntla: Now I'm confused with something even more basic. So context propagation uh user asks a question, user gets back an answer.
Rajashekar G: And
Yeshwanth Reddy Yerraguntla: Add a questioning and answering.
Rajashekar G: there.
Yeshwanth Reddy Yerraguntla: Okay. Uh see when you're giving back an ad hoc answer um that answer is going into visualization intelligence first thing.
Rajashekar G: Oh no.
Yeshwanth Reddy Yerraguntla: So visualization intelligence has the context of which user is talking and what preferences they
Rajashekar G: Oh,
Yeshwanth Reddy Yerraguntla: have. So while I agree that real time agent needs to know which
Rajashekar G: right.
Yeshwanth Reddy Yerraguntla: user is asking it doesn't matter because um when it generates a query when a data source or whichever whoever generates a query last week that when we maintain that uh DB level um I want a role level access for which user what rows are available on you will get back your answer and uh what do you call um the agent
 
 
02:18:43
 
Rajashekar G: Mhm.
Yeshwanth Reddy Yerraguntla: really doesn't need to know who asked it
Rajashekar G: Right.
Yeshwanth Reddy Yerraguntla: until u add query low that query is coming with the packet of user info Right?
Rajashekar G: Right.
Yeshwanth Reddy Yerraguntla: That packet has user info when which means when it makes a query on either the data agent or
Rajashekar G: Just
Yeshwanth Reddy Yerraguntla: the arc snapshot that user info anyway is injected into that query.
Rajashekar G: right.
Yeshwanth Reddy Yerraguntla: This is happening at a back end level not from agent perspective. agent key whatever information it is getting back it is always in the form of those rows which are open to the user only
Rajashekar G: Okay. My doubt is let's say I have
Yeshwanth Reddy Yerraguntla: yeah
Rajashekar G: a data uh real time intelligence agent question it should go and get it from the data agent so
Yeshwanth Reddy Yerraguntla: it will make an
Rajashekar G: data uh yes data agent multiple sources so now it should ask the uh exact agent let's say Salesforce agent it went to Salesforce agent and ask the question Salesforce agent needs to develop a SOQL query such that it
 
 
02:20:22
 
Yeshwanth Reddy Yerraguntla: H.
Rajashekar G: uh uh injected this governance also like which user is asking it and condition
Yeshwanth Reddy Yerraguntla: Yeah. Are the sales force set?
Rajashekar G: Salesforce headache Salesforce agent agent will give it to
Yeshwanth Reddy Yerraguntla: Yeah.
Rajashekar G: you agent
Yeshwanth Reddy Yerraguntla: Correct. So, real time agent is still not worrying about who asked it.
Rajashekar G: Right.
Yeshwanth Reddy Yerraguntla: Real time agent will simply say,
Rajashekar G: Just
Yeshwanth Reddy Yerraguntla: I I got this information. Now,
Rajashekar G: passing.
Yeshwanth Reddy Yerraguntla: I'll send it to UI agent. It will decide because who asked it how to show it?
Rajashekar G: Okay. And here comes my question. So output
Yeshwanth Reddy Yerraguntla: Where?
Rajashekar G: data uh has that governance rule or not
Yeshwanth Reddy Yerraguntla: No,
Rajashekar G: any
Yeshwanth Reddy Yerraguntla: but the fact that this data agent gave back an answer for that user itself means it was validated, right?
Rajashekar G: validation layer data agent. open
Yeshwanth Reddy Yerraguntla: H data isn't that validation layer has to be as close
 
 
02:21:39
 
Rajashekar G: like
Yeshwanth Reddy Yerraguntla: to the data source as possible. When you're sending the request,
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: you will send it along with uh which user is sending the request. So the data you'll always get back is for that user only trainer elevated permissions only full
Rajashekar G: Got it.
Yeshwanth Reddy Yerraguntla: data or refresh agent only refresh agent permissions are basically at full organization level. Right? So who is asking the question data source agent will know.
Rajashekar G: Right.
Yeshwanth Reddy Yerraguntla: So whatever answer it is giving should automatically be validated.
Rajashekar G: Right.
Yeshwanth Reddy Yerraguntla: We we will not separate validation from data response.
Rajashekar G: Got it. And one more note is like core and knowledge base is there
Yeshwanth Reddy Yerraguntla: Yeah.
Rajashekar G: and information or snapshot user information and three parts. So let's say if I having Salesforce, guit and insurance DB three connections are there. Three agents are also there three data agents. So whenever any question comes that will be passed to this data source agent and gets the information on the real time or are we duplicating the entire data into somewhere and fetching it from there
 
 
02:23:04
 
Yeshwanth Reddy Yerraguntla: Morning.
Rajashekar G: refresh we will be running a refresh at certain point and to get the latest snapshot while getting a snapshot are we getting the entire data Huh? Or we are only getting what needs to be get from that particular source.
Yeshwanth Reddy Yerraguntla: It's always deltas. So when agent says uh give me information from last 6 hours only that information agent will give only that information it will try to attach it to the
Rajashekar G: Mhm.
Yeshwanth Reddy Yerraguntla: existing graph.
Rajashekar G: Uh that will be like knowledge base but uh
Yeshwanth Reddy Yerraguntla: Correct?
Rajashekar G: entire
Yeshwanth Reddy Yerraguntla: Yeah.
Rajashekar G: duplicate.
Yeshwanth Reddy Yerraguntla: No that's uh we should not do that.
Rajashekar G: So let's say Salesforce is there. Right now our architecture is we are using simple Salesforce to directly connect to the Salesforce whenever it is
Yeshwanth Reddy Yerraguntla: H.
Rajashekar G: needed and to execute the queries on that. So to understand it in particular to Salesforce real time intelligence agent will talk to Salesforce agent whenever it is required
Yeshwanth Reddy Yerraguntla: Hm.
 
 
02:24:35
 
Rajashekar G: and that will generate a query and execute on the database Salesforce database
Yeshwanth Reddy Yerraguntla: Yes. Correct.
Rajashekar G: and refresh agent will have set of questions to get from the Salesforce. If we configure it like every six hours
Yeshwanth Reddy Yerraguntla: Yes.
Rajashekar G: agent again and get the information.
Yeshwanth Reddy Yerraguntla: Got it.
Rajashekar G: Okay. So in the entire process we are not duplicating the entire data anywhere. Okay.
Yeshwanth Reddy Yerraguntla: No.
Rajashekar G: This is fine for Salesforce but we come to guit that's a different
Yeshwanth Reddy Yerraguntla: Mhm. Correct.
Rajashekar G: mechanism
Yeshwanth Reddy Yerraguntla: when we are dealing with guit uh what do you mean different mechanism? Why is it
Rajashekar G: different mechanism means we need to have all the vector data in our
Yeshwanth Reddy Yerraguntla: different? Yes,
Rajashekar G: database.
Yeshwanth Reddy Yerraguntla: the data will come and sit here for on a day-to-day basis.
Rajashekar G: H
Yeshwanth Reddy Yerraguntla: whatever uh things have been happening I will store all of them in this rag DB for Gmail even though data source is Gmail we will still so
 
 
02:26:03
 
Rajashekar G: right
Yeshwanth Reddy Yerraguntla: it's not a near it's not a full duplicate it's sort of a duplicate in the sense that you have ids and uh the embedding uh vector followed by some additional indexes like what is the
Rajashekar G: vector
Yeshwanth Reddy Yerraguntla: sentiment of the email
Rajashekar G: Right.
Yeshwanth Reddy Yerraguntla: uh a vidanga you are not duplicating anything but when you have to send back the refresh information you'll send the full information saying uh I received these five emails first email belongs to this project in this project so and so discussions happened all those things you'll send back snapshot so the same Gmail is living in three different locations now in the original source in the Gmail email connector in the form of embeddings and finally in the form
Rajashekar G: Right.
Yeshwanth Reddy Yerraguntla: of orc snapshot.
Rajashekar G: Right.
Yeshwanth Reddy Yerraguntla: Each has different purpose.
Rajashekar G: Right. Uh here. So coming to the Gmail,
Yeshwanth Reddy Yerraguntla: other.
Rajashekar G: we won't get the exact real information at that point. Let's say if my refresh or sync is at a daily ones.
 
 
02:27:25
 
Rajashekar G: Uh some mails I got from morning 9 to 12.
Yeshwanth Reddy Yerraguntla: Yeah.
Rajashekar G: Those won't be in my database yet. If I ask anything about it, it won't know.
Yeshwanth Reddy Yerraguntla: That's why our refresh should be very high for certain connectors.
Rajashekar G: All
Yeshwanth Reddy Yerraguntla: Um yeah that said uh ammono if it
Rajashekar G: right.
Yeshwanth Reddy Yerraguntla: is uh yeah if it is soon uh sorry if it is not the right time the arc snapshot will not reflect it.
Rajashekar G: I don't know.
Yeshwanth Reddy Yerraguntla: uh but if the question is such that I can directly go and query Jira or SQL Salesforce I'll probably get back the answer even though it's not in arc snapshot um give me all the uh tasks that were finished in the last four hours on Gall the last refresh happened
Rajashekar G: Right.
Yeshwanth Reddy Yerraguntla: uh um I don't Five back five hours back one more
Rajashekar G: Six hours back.
Yeshwanth Reddy Yerraguntla: hour is there for refresh but now I'm asking give me all task completed in the
Rajashekar G: Right.
 
 
02:28:35
 
Yeshwanth Reddy Yerraguntla: last 4 hours so naturally it will go to Jira uh make a query and give me back four hours data only from the data source it will not say I don't have refreshed it yet you are asking a timesensitive
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: question so data will know how to do some time sensitive querying
Rajashekar G: Right.
Yeshwanth Reddy Yerraguntla: Standalone React context propagation ad hoc question answering user
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: snapshot snapshot information on there traveling packets dashboard user context core agent enriched context radius I don't know it's too early to say chat session so
Rajashekar G: session. We can remove it.
Yeshwanth Reddy Yerraguntla: manga what packet you'll receive is uh I will receive Give conversation history. Full conversation history also there. Then uh agent core agent will make a call saying uh it might make several calls. This is where actual branching happens, right?
Rajashekar G: You're right.
Yeshwanth Reddy Yerraguntla: and call either uh data agent or uh core uh knowledge base and uh synthesize answer back agent information and core agent will send back the information to UI agent.
 
 
02:30:46
 
Yeshwanth Reddy Yerraguntla: This is the
Rajashekar G: Hello
Yeshwanth Reddy Yerraguntla: path.
Rajashekar G: quickly.
Yeshwanth Reddy Yerraguntla: again. So this is essentially ad hoc question answering. Uh how much of it are we already doing in UI agent? UI agent is taking care of showing on the UI all those things. So what should this take care of when it comes to an MVP?
Rajashekar G: They should take care of orchestration
Yeshwanth Reddy Yerraguntla: Correct.
Rajashekar G: basically.
Yeshwanth Reddy Yerraguntla: Exactly. So they should uh MVP in two weeks. Features I have to include are what is that? Pantic AI orchestration next uh what else I am touching my I'm touching core knowledge base right related uh um tools what are those tools fetch Show or
Rajashekar G: Oh
Yeshwanth Reddy Yerraguntla: snapshot fetch runbooks run skill.
Rajashekar G: update. Graph update.
Yeshwanth Reddy Yerraguntla: update organization snapshot. Anyway, we are writing for the sake of completeness not uh any other agenda. So if there is a trainer who is saying uh okay I think training
 
 
02:32:51
 
Rajashekar G: I don't know.
Yeshwanth Reddy Yerraguntla: training life cycle update organization snapshot update runbooks update
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: skills uh core knowledge base related tools run I should also orchestration I
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: have uh data Uh, agent run tools or query tools.
Rajashekar G: Update runs.
Yeshwanth Reddy Yerraguntla: Okay. Add occurring low.
Rajashekar G: Then more.
Yeshwanth Reddy Yerraguntla: basic questions answers to answer it needs to hit data agents or its own knowledge base.
Rajashekar G: link that is coming to the
Yeshwanth Reddy Yerraguntla: Yeah.
Rajashekar G: user.
Yeshwanth Reddy Yerraguntla: I think I covered update runbook update or
Rajashekar G: Take in from the user step by
Yeshwanth Reddy Yerraguntla: H
Rajashekar G: step update
Yeshwanth Reddy Yerraguntla: center agent. So what is context
Rajashekar G: The only context propagation only session management.
Yeshwanth Reddy Yerraguntla: propagation?
Rajashekar G: We can pull few things down.
Yeshwanth Reddy Yerraguntla: Calc MVP the Pan let's not focus on that that column be empty it's okay news and insights Kalama.
Rajashekar G: News insights only real time relation agent is dealing with ad hoc queries and uh this
 
 
02:37:15
 
Yeshwanth Reddy Yerraguntla: Correct.
Rajashekar G: one.
Yeshwanth Reddy Yerraguntla: Huh? Correct. It is not touching the user aware inside distribution
Rajashekar G: Oh no.
Yeshwanth Reddy Yerraguntla: engine.
Rajashekar G: So this can come from refresh user
Yeshwanth Reddy Yerraguntla: Refresh agents necessarily.
Rajashekar G: interface.
Yeshwanth Reddy Yerraguntla: It will go to user aware agent refresh is refresh.
Rajashekar G: Hello
Yeshwanth Reddy Yerraguntla: Someone has to decide
Rajashekar G: diagram agent then that will process and give it to user aware insight and alert distribution engine.
Yeshwanth Reddy Yerraguntla: refresh information necessarily will go to user aware agent.
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: Ad hoc information will necessarily by bypass user aware of agent. Okay.
Rajashekar G: And I'm asking Amula to join this call also.
Yeshwanth Reddy Yerraguntla: Okay.
Rajashekar G: So she doesn't have other things at all.
Yeshwanth Reddy Yerraguntla: Yeah, sure. Color them.
Satyasri Prabhakar Mantripragada: We'll be dropping off now.
Yeshwanth Reddy Yerraguntla: Sure enough. Yeah.
Satyasri Prabhakar Mantripragada: Maybe after this session we should have that plan clear at least a
Yeshwanth Reddy Yerraguntla: Yeah.
 
 
02:39:03
 
Satyasri Prabhakar Mantripragada: stack rank list of priorities and then assign them to the
Yeshwanth Reddy Yerraguntla: Yes.
Satyasri Prabhakar Mantripragada: individuals.
Yeshwanth Reddy Yerraguntla: Yeah.
Satyasri Prabhakar Mantripragada: Every time we are complaining that we don't have direction or we don't have timelines now at least we have some uh bandwidth.
Yeshwanth Reddy Yerraguntla: Yeah. Got it. True. Yeah. Why?
Rajashekar G: Bice
Yeshwanth Reddy Yerraguntla: So color margin trigger refresh
Rajashekar G: know that's I
Yeshwanth Reddy Yerraguntla: refresh mechanism that will go to uh all the data agents. data agents will give back information that will go to insight uh agent that will send the broadcast
Rajashekar G: Right. Real time intelligence
Yeshwanth Reddy Yerraguntla: so
Rajashekar G: agentcess.
Yeshwanth Reddy Yerraguntla: e process real time will ping every guy that is one activity second activity is because it is getting information from all the agents it will have to do that uh Updating of arc snapshot.
Rajashekar G: Updating snapshots can be done by user aware
Yeshwanth Reddy Yerraguntla: No user aware insight is only doing these are the 100
 
 
02:40:33
 
Rajashekar G: inside.
Yeshwanth Reddy Yerraguntla: items the there are thousand people in my organization. How can I distribute all the news to every person in the organization? And
Rajashekar G: Okay. So maybe just confirm this. So real time refresh agent or refresh mechanism will be triggering a ping to real time dation intelligence agent. Real time dation intelligation agent is pinging to all the data source agent to fetch what what needed.
Yeshwanth Reddy Yerraguntla: Got it.
Rajashekar G: All the data source agent not giving the information back to real time dation intelligence.
Yeshwanth Reddy Yerraguntla: They are giving
Rajashekar G: Uh giving back to real time relation intelligence only that is updating the database or
Yeshwanth Reddy Yerraguntla: right.
Rajashekar G: snapshot uh or information update that will be triggering the user aware insight and alert distribution engine. Let's say some change happen in my database user aware insight and alert distribution
Yeshwanth Reddy Yerraguntla: Huh?
Rajashekar G: engine will be activated and triggering to the UI
Yeshwanth Reddy Yerraguntla: Correct.
Rajashekar G: connection like real time intelligence agent key insight alerts broadcast news.
 
 
02:42:17
 
Rajashekar G: Oh no. Okay.
Yeshwanth Reddy Yerraguntla: Next step is to call this guy with all the news.
Rajashekar G: Okay. Got it. So then logic same kind of questions related to data source agent. If our pit doesn't have anything now.
Yeshwanth Reddy Yerraguntla: Yeah, wherever if they whoever wants to join should join.
Rajashekar G: Yeah.
Yeshwanth Reddy Yerraguntla: Break out rooms concept
Rajashekar G: What's that?
Yeshwanth Reddy Yerraguntla: and Manisha they join. I was thinking they can go into a separate room, but they can always create a fresh melting for themselves.
Rajashekar G: Learning 1 hour, 1 hour morning.
Yeshwanth Reddy Yerraguntla: Yeah, I know. I know. I think uh breakout rooms.
Vara Kumar Jagarapu: You don't need a negative option.
Yeshwanth Reddy Yerraguntla: We have to enable that option or if I enable it, how can I start it here? Team your tools. Oh, sorry. Breakout rooms technically Manisha can up two
Vara Kumar Jagarapu: I think we need to drag or
Yeshwanth Reddy Yerraguntla: rooms.
 
 
02:44:38
 
Yeshwanth Reddy Yerraguntla: Timer shuffle clear breakout one.
Vara Kumar Jagarapu: what?
Yeshwanth Reddy Yerraguntla: Yeah, when the time comes we can always do it. I just wanted to explore. So
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: uh
Rajashekar G: Five minutes on message group. Okay.
Yeshwanth Reddy Yerraguntla: Okay. Sorry.
Rajashekar G: Yeah.
Yeshwanth Reddy Yerraguntla: 9.
Vara Kumar Jagarapu: Just one question more
Yeshwanth Reddy Yerraguntla: Yeah. Um,
Vara Kumar Jagarapu: if you can do it.
Yeshwanth Reddy Yerraguntla: check.
Rajashekar G: No or
Yeshwanth Reddy Yerraguntla: Current
Rajashekar G: Manisha.
Vara Kumar Jagarapu: I'll be having a lot of questions actually.
Yeshwanth Reddy Yerraguntla: current enterprise brain walk through just I think uh that should
Vara Kumar Jagarapu: I'll be
Yeshwanth Reddy Yerraguntla: answer right.
Vara Kumar Jagarapu: getting
Rajashekar G: So based on
Yeshwanth Reddy Yerraguntla: Yeah.
Rajashekar G: Yeah.
Vara Kumar Jagarapu: connect with them also.
Rajashekar G: Okay. Open
Vara Kumar Jagarapu: Sure. Okay.
Rajashekar G: time.
Vara Kumar Jagarapu: on multiple Google M. Okay. Sorry.
Yeshwanth Reddy Yerraguntla: know what?
Vara Kumar Jagarapu: Happy.
 
 
02:46:45
 
Rajashekar G: And then refresh agent
Yeshwanth Reddy Yerraguntla: Okay.
Rajashekar G: point.
Yeshwanth Reddy Yerraguntla: If people who want to edit it, uh just log into the uh system. That way you can edit anonymous.
Rajashekar G: H
Yeshwanth Reddy Yerraguntla: Those guys can't edit anything.
Rajashekar G: actually
Yeshwanth Reddy Yerraguntla: Okay. Okay.
Rajashekar G: window.
Yeshwanth Reddy Yerraguntla: Sir I am s you can take your
Vara Kumar Jagarapu: 10 15 minutes. I just need to
Yeshwanth Reddy Yerraguntla: time I think 40 45 minutes.
Vara Kumar Jagarapu: Okay.
Yeshwanth Reddy Yerraguntla: Yeah.
Vara Kumar Jagarapu: ready with them.
Yeshwanth Reddy Yerraguntla: Yeah.
Vara Kumar Jagarapu: Yeah.
Yeshwanth Reddy Yerraguntla: Check news life cycle. News life cycle
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: low.
Rajashekar G: Take it over. RTD. Yeah. Time reached.
Manisha Gundapuneedi: Hello.
Yeshwanth Reddy Yerraguntla: Hello.
Manisha Gundapuneedi: What just
Yeshwanth Reddy Yerraguntla: Cool.
Rajashekar G: Let's figure
Yeshwanth Reddy Yerraguntla: Data agent flow
Rajashekar G: out
Yeshwanth Reddy Yerraguntla: features.
Manisha Gundapuneedi: Okay,
Yeshwanth Reddy Yerraguntla: Uh so
Manisha Gundapuneedi: actually
Yeshwanth Reddy Yerraguntla: main
 
 
02:51:11
 
Rajashekar G: UA component.
Yeshwanth Reddy Yerraguntla: uh yeah Second wire.
Rajashekar G: Okay. Now
Yeshwanth Reddy Yerraguntla: So
Rajashekar G: I need
Manisha Gundapuneedi: And
Yeshwanth Reddy Yerraguntla: ammonilot
Manisha Gundapuneedi: it's
Rajashekar G: that.
Yeshwanth Reddy Yerraguntla: kit library react kit kit
Rajashekar G: Open it. G.
Manisha Gundapuneedi: It suggest
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: What I want is a UI, sorry, an application that acts very similar to Pylog. Um doc back end app agents executing agent pilgi.py Pilo there is this function something called app factory here. All it is taking is one pyantic AI agent, right? And um I like what that agent is doing, what that AI components are doing.
Rajashekar G: Yeah.
Yeshwanth Reddy Yerraguntla: E appactory canos all it is saying is I will act as a uh middleman between the user and the agent. User will ask questions. I will send it to agent. uh agent will give back answers. So I'll send I'll give put it back on the UI.
 
 
02:53:35
 
Yeshwanth Reddy Yerraguntla: So copilot kit to RD if we can come up with this kind of a shell with all these features all these features. Um basically that will be the starting point for our user interface. React.
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: So Kit
Rajashekar G: Enterprise ready front end stack for agents.
Yeshwanth Reddy Yerraguntla: Can I open?
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: Oh.
Rajashekar G: Open this.
Yeshwanth Reddy Yerraguntla: I think
Rajashekar G: system.
Yeshwanth Reddy Yerraguntla: probably. So there is this thing called genui generative UI. I want Anita to research on this
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: as what how is it working that given an agent uh uh it is able to generate uh dynamic UI on the fly. So um main challenge
Rajashekar G: Oh,
Yeshwanth Reddy Yerraguntla: is main challenge is to make this work with pantic because everything we are doing is in Python back end and this is react identity connect if she can do it that will be really great for today and tomorrow.
 
 
02:55:54
 
Rajashekar G: If Ganesh is free, he can also involve in this. He can guide us better.
Yeshwanth Reddy Yerraguntla: Yeah.
Rajashekar G: One more point we need to see is like there will be some animations. So Check capabilize
Manisha Gundapuneedi: All kinds of widgets.
Yeshwanth Reddy Yerraguntla: Got it.
Rajashekar G: 3D.
Manisha Gundapuneedi: Um,
Yeshwanth Reddy Yerraguntla: So
Rajashekar G: I'm going to
Yeshwanth Reddy Yerraguntla: library
Rajashekar G: go
Yeshwanth Reddy Yerraguntla: scripts examples. This
Rajashekar G: chat with identical
Yeshwanth Reddy Yerraguntla: IPL optin manufacturing IPL this IPL agent um is coming from this. Yeah. This is nothing but a plain IPL. um agent that can answer questions for IPL uh questions who run who scored most runs in 2017. It will make an SQL query and it'll give back an answer. If you can make this agent work with copilot kit uh that will be a good first step. So okay someone will run two um servers one is on the back end one is for the front
Rajashekar G: Right.
 
 
02:58:21
 
Yeshwanth Reddy Yerraguntla: end those two should talk to each other and front end when I ask a question this should they should get triggered
Rajashekar G: Better. I put the HS
Yeshwanth Reddy Yerraguntla: Yeah.
Rajashekar G: Wake
Yeshwanth Reddy Yerraguntla: Try
Rajashekar G: up.
Yeshwanth Reddy Yerraguntla: an
Rajashekar G: Anisha Kundi M.
Yeshwanth Reddy Yerraguntla: This is what we trying. We can review
Manisha Gundapuneedi: Ganesh
Rajashekar G: Simulator
Yeshwanth Reddy Yerraguntla: Anyway,
Manisha Gundapuneedi: Next,
Rajashekar G: Close.
Yeshwanth Reddy Yerraguntla: um
Manisha Gundapuneedi: Ganesh Rahul.
Yeshwanth Reddy Yerraguntla: Okay.
Manisha Gundapuneedi: Fullledge.
Yeshwanth Reddy Yerraguntla: H
Manisha Gundapuneedi: Massive testing.
Yeshwanth Reddy Yerraguntla: Okay.
Manisha Gundapuneedi: Continue
Yeshwanth Reddy Yerraguntla: Office
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: calendar.
Manisha Gundapuneedi: discussion.
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: Sorry. Record.
Manisha Gundapuneedi: Okay. Okay.
Yeshwanth Reddy Yerraguntla: Hello.
Arpit Pathak: Happy only.
Yeshwanth Reddy Yerraguntla: Holiday.
Arpit Pathak: Sadly,
Yeshwanth Reddy Yerraguntla: What? So
Arpit Pathak: no.
Yeshwanth Reddy Yerraguntla: weird.
Arpit Pathak: You're not going there.
Yeshwanth Reddy Yerraguntla: I don't play in India also.
Arpit Pathak: Okay. So no ta also
 
 
03:14:01
 
Yeshwanth Reddy Yerraguntla: No.
Arpit Pathak: nothing you are antony.
Yeshwanth Reddy Yerraguntla: You don't I never liked it.
Arpit Pathak: Yeah, it's actually played in North I have few doubts regarding architecture.
Yeshwanth Reddy Yerraguntla: Yeah.
Arpit Pathak: I sharing the screen.
Yeshwanth Reddy Yerraguntla: Yeah.
Arpit Pathak: So I I was going through this data source area. Uh I have these points uh to get clarified.
Yeshwanth Reddy Yerraguntla: Yeah.
Arpit Pathak: Uh so one was run books versus skills. So in optin what we did was run book was the setup instructions that we are given to
Yeshwanth Reddy Yerraguntla: Huh?
Arpit Pathak: agent uh including a tools. Uh but here there are two explicit terms uh books and skills.
Yeshwanth Reddy Yerraguntla: Yeah, I am also still not clear on how to differentiate this but
Arpit Pathak: Uhhuh.
Yeshwanth Reddy Yerraguntla: mainly I mean it can so happen that both of them will get combined into one thing only eventually.
Arpit Pathak: Okay, I'll put it in research then for now.
Yeshwanth Reddy Yerraguntla: So I will suggest you should give higher priority to skills because skills is trying to cover both the concepts that skills is a set of
 
 
03:15:43
 
Arpit Pathak: Okay.
Yeshwanth Reddy Yerraguntla: instructions and uh set of
Arpit Pathak: Tools uh functionalities as in
Yeshwanth Reddy Yerraguntla: functionalities.
Arpit Pathak: tools.
Yeshwanth Reddy Yerraguntla: Correct. Correct. Correct. Yeah.
Arpit Pathak: So can I consider the optins run books as baseline?
Yeshwanth Reddy Yerraguntla: Can I consider optins on books as skills?
Arpit Pathak: Yeah.
Yeshwanth Reddy Yerraguntla: That's what to an extent uh the answer is
Arpit Pathak: Okay.
Yeshwanth Reddy Yerraguntla: yes.
Arpit Pathak: Sounds good. Uh second point was update and knowledge. So for now I'm assuming that knowledge is static. Uh we what we are doing is we are chunking the parts and then storing into the editor. All these things are and then we do semantic search. Uh but I have one concern because we are dealing with the data sources. So data will get modified too in that in
Yeshwanth Reddy Yerraguntla: M okay I mean the source data you're
Arpit Pathak: that case uh
Yeshwanth Reddy Yerraguntla: saying what do you mean
Arpit Pathak: sorry I mean for
Yeshwanth Reddy Yerraguntla: data gets modified
 
 
03:17:00
 
Arpit Pathak: example let's say today uh sales force data shows something something escalation due to XYZ factor Would that also stored in knowled
Yeshwanth Reddy Yerraguntla: Those are mainly what are uh supposed to be treated as uh that is in my head they were run books.
Arpit Pathak: so they will immediately get fetched and got uh executed.
Yeshwanth Reddy Yerraguntla: Got
Arpit Pathak: So here we mentioned rag DB but on the rag DB we are mentioning runbooks.
Yeshwanth Reddy Yerraguntla: it.
Arpit Pathak: Run books can be modified too but if we are considering rag I'm putting it as rag then updating in the rag and then again fing it would it be insufficient.
Yeshwanth Reddy Yerraguntla: Uh, repeat that.
Arpit Pathak: Uh so right now we are considering runbook skills in rag DB
Yeshwanth Reddy Yerraguntla: I mean it's written for the sake of writing it. It's not like these will have their own
Arpit Pathak: okay that's that point I was thinking
Yeshwanth Reddy Yerraguntla: embeddings. It's a hybrid DB where rag is also available.
Arpit Pathak: yesterday okay
Yeshwanth Reddy Yerraguntla: That's how you should look at
 
 
03:18:23
 
Arpit Pathak: okay
Yeshwanth Reddy Yerraguntla: it.
Arpit Pathak: got it so essentially these could be separate also
Yeshwanth Reddy Yerraguntla: Huh? Postgress may you can have multiple databases, multiple
Arpit Pathak: Mhm.
Yeshwanth Reddy Yerraguntla: tables.
Arpit Pathak: Okay. Okay. So, DV mentioned that. Okay. DV was the key vector. RAG is one of them.
Yeshwanth Reddy Yerraguntla: Yes.
Arpit Pathak: Uh then okay. So, this point got cleared. Uh so yeah one more thing right now in the current application they are using red cache uh and
Yeshwanth Reddy Yerraguntla: I repeat.
Arpit Pathak: huh sorry okay so
Yeshwanth Reddy Yerraguntla: Sorry, I missed your voice. What are you saying?
Arpit Pathak: right now in the current uh application enterprise we are using rack caching uh and there is a rack so I was little confused regarding it. Uh that rack cache and rag I mean what would be yeah I mean
Yeshwanth Reddy Yerraguntla: H okay what is the difference
Arpit Pathak: a difference as in I got as a memory thing but still
Yeshwanth Reddy Yerraguntla: okay I'll tell you rag cache is like more like question answer kh with rag uh as the lookup
 
 
03:19:55
 
Arpit Pathak: Mhm.
Yeshwanth Reddy Yerraguntla: mechanism. So if if you are asking give me the top 10 uh uh you know
Arpit Pathak: Okay.
Yeshwanth Reddy Yerraguntla: run scorers in world cup and I am asking give me the
Arpit Pathak: Yeah.
Yeshwanth Reddy Yerraguntla: highest 10 uh uh run getters in this world cup both are same questions. So ideally if you have a rag enabled uh database where for your question
Arpit Pathak: Yeah.
Yeshwanth Reddy Yerraguntla: it has been uh uh you have the question for that question there is the embedding and for
Arpit Pathak: Mhm.
Yeshwanth Reddy Yerraguntla: that question you also have the SQL query associated with it.
Arpit Pathak: Okay.
Yeshwanth Reddy Yerraguntla: What then happens is my question will have a high match with that uh embedding. So it will directly fetch the SQL query and run it instead of generating the query from scratch.
Arpit Pathak: Okay. Okay. So, essentially when we are let's say I saw the chatting and how it is going to react. So let's say we are chatting in a the chat box.
 
 
03:20:59
 
Arpit Pathak: Okay. Uh we ask some questions and then I think there's a uh mode or something that we say okay do you want to reuse it or something then it goes to ra so that that should be used in that session right chat
Yeshwanth Reddy Yerraguntla: Correct.
Arpit Pathak: session.
Yeshwanth Reddy Yerraguntla: Correct.
Arpit Pathak: Uh so it it is session based
Yeshwanth Reddy Yerraguntla: H It is session based in the sense for okay
Arpit Pathak: session based
Yeshwanth Reddy Yerraguntla: uh why do you say session based like and
Arpit Pathak: because
Yeshwanth Reddy Yerraguntla: what I'm asking of the different sessions still I'm saying rag will be available
Arpit Pathak: okay rag will be available but the possibility of matching it uh
Yeshwanth Reddy Yerraguntla: Yeah.
Arpit Pathak: would be same for that session. Right? If you change the session and then ask the question there differently then it should ideally search for whole rag and once the conversation is built and then it should search for cash
Yeshwanth Reddy Yerraguntla: Correct. Correct.
Arpit Pathak: rather than main test
Yeshwanth Reddy Yerraguntla: Correct.
Arpit Pathak: ra should Okay.
 
 
03:22:20
 
Arpit Pathak: Uh starting with red. Yeah. So in in this way we can incorporate that thing of high matching. This part is done.
Yeshwanth Reddy Yerraguntla: All
Arpit Pathak: Uh yeah one more part uh yesterday I was talking to P uh regarding that uh data security. So he mentioned that we are going to use URLs rather than the raw text.
Yeshwanth Reddy Yerraguntla: right.
Arpit Pathak: So on URL we are going to uh we are going to store it in DB but at the same time we should also load that into rack as
Yeshwanth Reddy Yerraguntla: Um it depends on use case.
Arpit Pathak: right.
Yeshwanth Reddy Yerraguntla: So uh for Gmail yes you will load all your emails into ra
Arpit Pathak: Okay.
Yeshwanth Reddy Yerraguntla: cache.
Arpit Pathak: One more thing I think they are maintaining one table per bracket.
Yeshwanth Reddy Yerraguntla: I mean that is v_sub1. This is V2.
Arpit Pathak: So ideally shouldn't it shouldn't it be erased uh when a new session start or something like that to get a high match
Yeshwanth Reddy Yerraguntla: Shouldn't it be
 
 
03:24:35
 
Arpit Pathak: or yeah not or maybe I don't know rag because
Yeshwanth Reddy Yerraguntla: erased?
Arpit Pathak: rag is maintained anyway as a separate entity or how should
Yeshwanth Reddy Yerraguntla: like you want to
Arpit Pathak: it for example uh the context switching happened
Yeshwanth Reddy Yerraguntla: elaborate
Arpit Pathak: between two sess two sessions uh then both things will be
Yeshwanth Reddy Yerraguntla: right
Arpit Pathak: there. So this new brackets will be building up and the previous one will also be there.
Yeshwanth Reddy Yerraguntla: both of them are happening in the same DB I'm saying across all users also it's
Arpit Pathak: Oh same DB.
Yeshwanth Reddy Yerraguntla: the same DB that's my example now You asked for top 10
Arpit Pathak: Okay.
Yeshwanth Reddy Yerraguntla: players. I'm also asking for top 10.
Arpit Pathak: Mhm.
Yeshwanth Reddy Yerraguntla: How will it know that there is already an answer? Someone already asked this question. So, it's across sessions,
Arpit Pathak: Yeah.
Yeshwanth Reddy Yerraguntla: across users. It's a single source of truth.
Arpit Pathak: Okay. But answers will be different.
 
 
03:25:43
 
Yeshwanth Reddy Yerraguntla: It can be different like different based on different people's uh uh access to the databases. Uh yeah,
Arpit Pathak: Uhhuh.
Yeshwanth Reddy Yerraguntla: it might fetch different rows
Arpit Pathak: So answers are also going to set in D right.
Yeshwanth Reddy Yerraguntla: answers in the form of queries.
Arpit Pathak: Okay. Number of queries. Okay. So that would be helpful for further analysis. Okay.
Yeshwanth Reddy Yerraguntla: Yeah.
Arpit Pathak: Because answers will keep on changing once it uh data got increased. Yes.
Yeshwanth Reddy Yerraguntla: Yes.
Arpit Pathak: One more thing I I forgot about active. Active and inactive.
Yeshwanth Reddy Yerraguntla: Active. Inactive is just an example. You don't need to focus on it.
Arpit Pathak: Okay. Okay. I I'm clear at this point.
Yeshwanth Reddy Yerraguntla: Um,
Arpit Pathak: one. Okay. Okay son, I think I'm clear. I'll read about it all. Thank
Yeshwanth Reddy Yerraguntla: Yeah.
Rajashekar G: And now and I have added
Yeshwanth Reddy Yerraguntla: Har checker.
 
 
03:31:47
 
Rajashekar G: points for refresh agent life cycle.
Yeshwanth Reddy Yerraguntla: Okay.
Rajashekar G: Next alert.
Yeshwanth Reddy Yerraguntla: Sure.
Rajashekar G: So,
Yeshwanth Reddy Yerraguntla: Mhm.
Rajashekar G: so two ways like alert should come when data changed in data source let's say at night
Yeshwanth Reddy Yerraguntla: H.
Rajashekar G: it's around everyday 12 a.m. So 12
Yeshwanth Reddy Yerraguntla: Yeah.
Rajashekar G: a.m.
Yeshwanth Reddy Yerraguntla: Uh refresh. Yeah. Oh no. Oh no.
Rajashekar G: So we are we are solving this problem by making sure that refresh thing should happen for
Yeshwanth Reddy Yerraguntla: Correct.
Rajashekar G: this source. Okay. Let's say kind of sources real time. That number is updating in the database%.
Yeshwanth Reddy Yerraguntla: got reflect
Rajashekar G: So it should be continuously monitoring the database like every 10 minutes every 1 minute. If that solves it
Yeshwanth Reddy Yerraguntla: Um so the separate
Rajashekar G: all.
Yeshwanth Reddy Yerraguntla: mechanism Rascal one ideally what should happen there is there is no refresh mechanism dependency either data source agent will continuously listen to whatever it has to pay attention to.
 
 
03:34:16
 
Yeshwanth Reddy Yerraguntla: The moment some threshold hits or something worth escalating is happening, it will immediately ping the core agent saying this is something that needs to be alerted. Core agent will immediately then send it. So there is no time or polling in this cycle.
Rajashekar G: I repeat.
Yeshwanth Reddy Yerraguntla: screen share optionally when such high critical agents are there where I have to get information on
Rajashekar G: Mhm.
Yeshwanth Reddy Yerraguntla: at moment's notice. This will now constantly this data source agent will have there has to be some mechanism. We don't know what that mechanism is. Uh data access user.
Rajashekar G: Very
Yeshwanth Reddy Yerraguntla: Uh
Rajashekar G: cool.
Yeshwanth Reddy Yerraguntla: I'll get rid of this. Generally some databases allow for uh push requests uh sorry allow for pulling radius databases
Rajashekar G: push the
Yeshwanth Reddy Yerraguntla: slow support uh um uh pull mechanisms where some information changes immediately that is sent to the listener publish subscribe concept So in those cases where such a mechanism is available data itself
 
 
03:36:06
 
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: will push some alert to the data source agent something is important and it
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: sorry itself will
Rajashekar G: Data sourcent will directly
Yeshwanth Reddy Yerraguntla: send to data sorry data source will itself will send alert Pull push alert push notification. There
Rajashekar G: Some column key there will be some push
Yeshwanth Reddy Yerraguntla: are
Rajashekar G: alert.
Yeshwanth Reddy Yerraguntla: exactly there are databases which support push notifications all situation when you push uh send the push alert immediately that will send back to the main agent uh that there is an escalation.
Rajashekar G: Okay. Data source
Yeshwanth Reddy Yerraguntla: H that is only if the data source supports
Rajashekar G: level.
Yeshwanth Reddy Yerraguntla: it
Rajashekar G: Okay. Otherwise it will be run from like uh refresh
Yeshwanth Reddy Yerraguntla: otherwise you have to pull.
Rajashekar G: agent.
Yeshwanth Reddy Yerraguntla: Yeah, you have to poll.
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: I am thinking polling. Should it happen at a data source to source level? I don't
Rajashekar G: source will directly
 
 
03:37:40
 
Yeshwanth Reddy Yerraguntla: know.
Rajashekar G: give.
Yeshwanth Reddy Yerraguntla: S I don't think we need to discuss that part. Either way, it's a small activity only.
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: Alert life cycle.
Rajashekar G: Hello.
Yeshwanth Reddy Yerraguntla: Okay. Refresh. No, no. Configure setup of data source is needs to be fetched. Yeah. Ping at every refresh point. Yeah. User preference life cycle low. Manisha
Manisha Gundapuneedi: I
Yeshwanth Reddy Yerraguntla: RJ
Manisha Gundapuneedi: know.
Yeshwanth Reddy Yerraguntla: for features. No.
Manisha Gundapuneedi: Yes. Get out.
Yeshwanth Reddy Yerraguntla: Thank you. Byebye.
Rajashekar G: We just have fun. I make it down. Hello.
Yeshwanth Reddy Yerraguntla: All right. Check.
Rajashekar G: Alerts user insight alert distribution
Yeshwanth Reddy Yerraguntla: No.
Rajashekar G: alerts.
Yeshwanth Reddy Yerraguntla: No.
Rajashekar G: Any question answer
Yeshwanth Reddy Yerraguntla: Got it.
Manisha Gundapuneedi: A2A authenticated hardwood path with a single data source connector and evidence fails.
 
 
03:46:57
 
Rajashekar G: is the alert unless agent unless
Yeshwanth Reddy Yerraguntla: product.
Rajashekar G: like let's say okay insight alerts broadcast news every time it is giving to user aware insight and alert
Yeshwanth Reddy Yerraguntla: Yeah.
Rajashekar G: distribution based on that it is sending the info to UI it is not
Yeshwanth Reddy Yerraguntla: Yeah,
Rajashekar G: modifying
Yeshwanth Reddy Yerraguntla: whether it is modifying or not, you have to I mean what it is sending is what it thinks is the right thing to send. So it is modifying I mean it is receiving news
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: from data source
Rajashekar G: Right.
Yeshwanth Reddy Yerraguntla: agents.
Rajashekar G: Data source agents to RDDIA then that will be to user aware uh engine.
Yeshwanth Reddy Yerraguntla: Correct.
Rajashekar G: So three layers basically.
Yeshwanth Reddy Yerraguntla: Correct.
Rajashekar G: Okay. Now just Time logging.
Yeshwanth Reddy Yerraguntla: Yeah. Yeah. Yeah. Enjoy pasta.
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: I'm back.
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: Great work. It is thorough. A few observations.
 
 
03:56:50
 
Manisha Gundapuneedi: Anna
Yeshwanth Reddy Yerraguntla: Ch mania.
Manisha Gundapuneedi: refreshs.
Yeshwanth Reddy Yerraguntla: Choose
Rajashekar G: Refresh insert life cycle.
Yeshwanth Reddy Yerraguntla: refresh insights life cycle. H this is not okay.
Manisha Gundapuneedi: Secondly,
Yeshwanth Reddy Yerraguntla: Oh cuz I cannot draw. No. Okay. No.
Rajashekar G: Don't
Manisha Gundapuneedi: Input 13.
Yeshwanth Reddy Yerraguntla: Uh, so manual refresh API endoint.
Manisha Gundapuneedi: create ah Information agent information
Yeshwanth Reddy Yerraguntla: agent
Manisha Gundapuneedi: agent.
Rajashekar G: What's a question
Yeshwanth Reddy Yerraguntla: there's no API agent to agent
Rajashekar G: on
Manisha Gundapuneedi: Okay.
Yeshwanth Reddy Yerraguntla: communication then
Manisha Gundapuneedi: Okay. That is a song.
Yeshwanth Reddy Yerraguntla: um once you have the uh so basically what you have to write is uh enable uh uh receiving information from uh core agent actually that is the communication layer that uh is in the present past row. So only refresh specific task same that I will receive a refresh ping then I will
Rajashekar G: It's a good weather.
Yeshwanth Reddy Yerraguntla: uh uh incremental
 
 
04:01:14
 
Rajashekar G: Let's see.
Yeshwanth Reddy Yerraguntla: data pool logic uh
Manisha Gundapuneedi: last.
Yeshwanth Reddy Yerraguntla: oh my
Manisha Gundapuneedi: Okay. Duplicate. I mean duplicate.
Yeshwanth Reddy Yerraguntla: Duplicate refresh data agent. See core agent might ask core agent might ask uh uh give me fresh data from the last 6 hours. So this agent will um give 6 hours worth of data. Um derive insights, publish, refresh insights to RP agent. So six first thing is it should receive uh refresh information from core agent. That information should contain starting time and ending time. Then your agent will enable um what do you call increment data pull logic agent. Once the data is pulled, next step is to update its own database agent database. First update then
Manisha Gundapuneedi: only.
Yeshwanth Reddy Yerraguntla: it will send to the core agent. These are all the updates. So one action point is increment data pool logic. Second increment data pool logic in the next action point is uh next action point is save to DB send in news or broadcast send news or insights to core agent insight agent snapshot
 
 
04:03:43
 
Manisha Gundapuneedi: A1 screen share follows.
Yeshwanth Reddy Yerraguntla: Refresh ping receive refresh. Yes. Prevent duplicate refresh.
Manisha Gundapuneedi: already
Yeshwanth Reddy Yerraguntla: Okay, second.
Manisha Gundapuneedi: This
Rajashekar G: But it is not like refreshing. It is just agent data source agent. It is like only syncing the data again.
Manisha Gundapuneedi: I'll
Rajashekar G: If we are asking one question or multiple
Manisha Gundapuneedi: do
Rajashekar G: questions,
Yeshwanth Reddy Yerraguntla: correct the I mean it's an important point here. All you're trying to say is potency that if even if the core agent is asking 10 times, it will do only once. Sure. Run for every agree time by the
Rajashekar G: Right.
Yeshwanth Reddy Yerraguntla: way. And all that headache is taken care of core agent and uh the spelling mistake hon
Manisha Gundapuneedi: He's out of the oven.
Yeshwanth Reddy Yerraguntla: increment data pool logic then update uh DB self DB and indexes Gmail agent I will update all my u emails with embeddings and I'll also update sentiment for every email.
 
 
04:06:30
 
Yeshwanth Reddy Yerraguntla: That is the self update DB snapshot on emulator. Did I write anything like a snapshot
Rajashekar G: Snapchat. I'm going to an
Yeshwanth Reddy Yerraguntla: snapshot overall knowledge organization snapshot deriving insights? Uh you can yeah you can technically say these are the insights and publish refresh insights to the RT agent. Yes, this is correct. Allah um refresh escalation life cycle law
Manisha Gundapuneedi: Okay.
Yeshwanth Reddy Yerraguntla: most of it is same as above except during publishing uh refresh insights. So if most of it is same actually okay Publish refresh insights to RT along with evidence whatever it's giving it should give us some this is the core mail whatever source links same thing it could publish uh refresh alerts to RT agent in a different packet and when it is sending a JSON it will send two different keys on M one key will be insights one key will be alerts so refresh insights and escalation life cycle 90% same work we don't need to create two tasks training and learning life cycle agent behavior context life cycle Secure transport
 
 
04:08:32
 
Manisha Gundapuneedi: Ender.
Yeshwanth Reddy Yerraguntla: transport. This is the core mechanism by which two agents are talking to each other. Uh features pantic communication between real time RTD intelligence agent. RTDIA and data source agents both are different two different pya agents that they have to talk to each other and the only I mean the standard layer is a toa identic dog there is a communication So once you have an agent, you can simply run A2 to A and it will open up a uh port uh uh at 8,000 or whatever saying uh my agent is uh listening to other agents uh protocol HL1 that is what this uh using o and A2A protocol. Next we are saying add a query life cycle. Uh intelligence agent is asking a question. Data source agent will ping data source and get back answer. on authenticated path with single data source connector evidence B response features law um receive question life cycle receive question check against uh against rag cache Umh.
 
 
04:10:51
 
Rajashekar G: Rag Q question and answers. Okay.
Yeshwanth Reddy Yerraguntla: If yes,
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: fetch from cash A. If no, fetch from. If no, fetch uh uh best best practices run book. fetch the best skill in generate query. Uh save in QA cache and finally respond back with answer. Right? So the this is ad hoc life cycle. training lesson. What I'll suggest is uh once you have all these features definitely is this the right direction he
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: can also help us.
Manisha Gundapuneedi: Okay.
Rajashekar G: Shall we pull him once here in the area? Okay.
Yeshwanth Reddy Yerraguntla: Okay. Uh you can whatever Excel prioritize JC all those things you can do.
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: Um yeah now that you'll have enough items we can prioritize. Anyone
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: questions?
Rajashekar G: validation.
Yeshwanth Reddy Yerraguntla: I also agree. Huh? P the
Rajashekar G: Okay. Okay.
Yeshwanth Reddy Yerraguntla: Yeah.
Rajashekar G: Okay. Nothing. Enterprise brain
Manisha Gundapuneedi: Ah.
Rajashekar G: enterprise brain and visual effect detection and enterprise for manufacturing.
Yeshwanth Reddy Yerraguntla: Yeah. Yeah. Yeah. Perfect. On there. Yeah. Yeah.
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: Yeah. Yeah.
Rajashekar G: Okay.
Yeshwanth Reddy Yerraguntla: Drop.
Manisha Gundapuneedi: Okay.
Rajashekar G: eyes
Yeshwanth Reddy Yerraguntla: Yeah. Uh,
Manisha Gundapuneedi: Hello.
Yeshwanth Reddy Yerraguntla: bro.
Rajashekar G: here.
Yeshwanth Reddy Yerraguntla: You can see that link validate.
Vara Kumar Jagarapu: Okay. Okay. Yeah. Afternoon.
Rajashekar G: Oh yeah.
Vara Kumar Jagarapu: Sorry.
Rajashekar G: Okay. Thank you.
Vara Kumar Jagarapu: Uh
Rajashekar G: Fine. Okay.
Vara Kumar Jagarapu: okay. Just walk through questions.
 
 
Transcription ended after 04:17:06

This editable transcript was computer generated and might contain errors. People can also change the text after it was created.