Meeting Title: Eden AI Project Brainstorming Date: 2026-04-13 Meeting participants: Pranav Narahari, Awaish Kumar
WEBVTT
1 00:02:50.590 ⇒ 00:02:52.190 Awaish Kumar: Hi, Panel.
2 00:02:57.220 ⇒ 00:02:58.150 Awaish Kumar: Hello?
3 00:02:59.760 ⇒ 00:03:00.630 Pranav Narahari: Hey, Wish.
4 00:03:02.820 ⇒ 00:03:03.879 Awaish Kumar: Hi, how you doing?
5 00:03:04.390 ⇒ 00:03:06.149 Pranav Narahari: Pretty good, pretty good, how you doing?
6 00:03:08.940 ⇒ 00:03:09.770 Awaish Kumar: Okay.
7 00:03:11.550 ⇒ 00:03:13.500 Pranav Narahari: Sorry, it cut out, what was that?
8 00:03:15.440 ⇒ 00:03:18.610 Awaish Kumar: I think that’s… Okay, yeah.
9 00:03:20.990 ⇒ 00:03:22.019 Pranav Narahari: Okay, can you hear me?
10 00:03:22.020 ⇒ 00:03:29.109 Awaish Kumar: I just wanted to… yeah, I just wanted to express the… The Eden AI project?
11 00:03:29.960 ⇒ 00:03:34.069 Pranav Narahari: Yes, yeah, I just don’.
12 00:03:34.070 ⇒ 00:03:34.510 Awaish Kumar: for college.
13 00:03:34.510 ⇒ 00:03:35.190 Pranav Narahari: Danny?
14 00:03:35.350 ⇒ 00:03:43.669 Pranav Narahari: I know that, yeah, Utam sent over that message, this is…
15 00:03:44.580 ⇒ 00:03:49.400 Pranav Narahari: the… I think with these prompts, right.
16 00:03:49.540 ⇒ 00:03:52.230 Pranav Narahari: it’ll make a little bit more sense, and I see…
17 00:03:52.600 ⇒ 00:04:01.619 Pranav Narahari: I think initially, right, we were thinking we’ll stand up a pretty, like, a database, we’ll have, like, an ETL job, all these things. Then…
18 00:04:01.940 ⇒ 00:04:08.779 Pranav Narahari: There was talk about using GWS CLI to pull in the information at request time.
19 00:04:09.030 ⇒ 00:04:14.000 Pranav Narahari: Nope. What was that? Sorry.
20 00:04:15.270 ⇒ 00:04:18.609 Awaish Kumar: Yeah, yeah, I’m just saying yes, I’m… Oh, okay.
21 00:04:18.909 ⇒ 00:04:23.939 Pranav Narahari: Perfect, yeah, yeah. But… now I think…
22 00:04:24.639 ⇒ 00:04:29.319 Pranav Narahari: Utam’s concern is that we cannot do that.
23 00:04:30.669 ⇒ 00:04:45.889 Pranav Narahari: Now, I kinda see what he’s saying, especially after my… well, I don’t… I don’t even wanna… because I don’t want to be the one to dictate the exactual… exact technical design.
24 00:04:47.319 ⇒ 00:04:55.859 Pranav Narahari: However, I do have, like, some ideas, and I’m gonna bring to you exactly what the client wants to use this product for, right?
25 00:04:56.279 ⇒ 00:04:58.049 Pranav Narahari: And so, what… yeah.
26 00:04:58.050 ⇒ 00:05:00.090 Awaish Kumar: Yeah, so there are, like.
27 00:05:00.380 ⇒ 00:05:06.550 Awaish Kumar: two aspects, right? One is AI part, and then there is a data part, right?
28 00:05:06.700 ⇒ 00:05:13.030 Awaish Kumar: So when I… so that’s why I was relying more on… on both of you guys regarding…
29 00:05:13.450 ⇒ 00:05:18.169 Awaish Kumar: how the AI or LLM, whatever you’re trying to use.
30 00:05:18.910 ⇒ 00:05:21.109 Awaish Kumar: Going to read the data from, right?
31 00:05:21.320 ⇒ 00:05:25.699 Awaish Kumar: So… like, I have worked before with the…
32 00:05:25.840 ⇒ 00:05:30.100 Awaish Kumar: Casey, Mustafa, and one other guy, there was.
33 00:05:30.250 ⇒ 00:05:33.900 Awaish Kumar: So, when we were creating these slackboards, right?
34 00:05:35.530 ⇒ 00:05:38.669 Awaish Kumar: So I helped them build this super-based database.
35 00:05:38.960 ⇒ 00:05:43.510 Awaish Kumar: in that superbase, recreate… we… I actually…
36 00:05:45.180 ⇒ 00:05:51.469 Awaish Kumar: get all the data from all select channels, and dump it in a Superbase repository.
37 00:05:51.610 ⇒ 00:05:59.009 Awaish Kumar: And where they basically… They used to, for the documents, especially, they used to create
38 00:05:59.210 ⇒ 00:06:04.040 Awaish Kumar: these RAG, like, encodings and… And save it in a table.
39 00:06:04.930 ⇒ 00:06:13.470 Awaish Kumar: Right? And then they basically… the search, like… and then basically that is being used. So I was just, like.
40 00:06:13.700 ⇒ 00:06:22.499 Awaish Kumar: Sending new data, to those repositories, and they will just create a RAG, like, kind of pre-processing.
41 00:06:22.770 ⇒ 00:06:25.079 Awaish Kumar: step, that was, like, so these…
42 00:06:25.450 ⇒ 00:06:28.589 Awaish Kumar: like, because the data is staring. Once it is in there, like.
43 00:06:28.690 ⇒ 00:06:33.020 Awaish Kumar: It’s not changing, right? If it’s a Slack message, email, or anything.
44 00:06:33.290 ⇒ 00:06:39.579 Awaish Kumar: There might be updates, but the existing data isn’t changing, so wherever data comes in, we create these
45 00:06:39.920 ⇒ 00:06:45.780 Awaish Kumar: RAG indexing and encodings, and then… they basically…
46 00:06:46.460 ⇒ 00:06:50.450 Awaish Kumar: use those… that as an input for NLMs, and then…
47 00:06:50.800 ⇒ 00:06:55.510 Awaish Kumar: I don’t exactly know the AI part, so that’s why I might be… mixture.
48 00:06:55.510 ⇒ 00:07:08.739 Pranav Narahari: Yeah, so let me kind of tell you what I think is necessary from a data part, right? And I can save the conversation for AI with maybe Sam, and then the three of us can probably just talk at the end, just make sure the end-to-end everything makes sense.
49 00:07:09.530 ⇒ 00:07:15.079 Awaish Kumar: I’m just giving an example, because this gives me, like, clarity. Like, when I sit with them.
50 00:07:15.290 ⇒ 00:07:26.029 Awaish Kumar: sat with them, like, on this… on this project, they were like, okay, we need this data, we need to do RAG, we need to use RAG, and then we need to basically
51 00:07:26.170 ⇒ 00:07:30.230 Awaish Kumar: Like, vectorize the query, and then…
52 00:07:30.500 ⇒ 00:07:41.269 Awaish Kumar: get the… based on similarity, get the response, right? So that was the project, and… and then it was clear, like, they need a database where they can store the data, they needed…
53 00:07:41.500 ⇒ 00:07:48.750 Awaish Kumar: maybe Slack conversation as a TXT files, so I was, like, writing ETL pipelines to bring that data in.
54 00:07:49.130 ⇒ 00:07:51.109 Awaish Kumar: From… oh, yeah, from now… from.
55 00:07:51.110 ⇒ 00:07:54.250 Pranav Narahari: I see what you’re saying. Yeah, so, kind of…
56 00:07:54.930 ⇒ 00:08:04.950 Pranav Narahari: Sam and… or, Casey Mustafa came to you, kind of, with, like, hey, we want to do… use a rag design. Now, tell me the process for doing that.
57 00:08:05.080 ⇒ 00:08:10.909 Pranav Narahari: See, now, the problem with that is sometimes rag design is just not the right choice.
58 00:08:11.060 ⇒ 00:08:19.500 Pranav Narahari: Right? Sometimes you can actually just have the CLI pull the information ad hoc. You don’t need to actually have, like, all the embeddings
59 00:08:19.660 ⇒ 00:08:36.110 Pranav Narahari: pre-created. And so, like, not saying that for that example, that’s how it should have went, I’m just saying for… we need to see a little bit higher level before we kind of dig into the exact implementation. So, like, RAG is already a decision, right? I don’t…
60 00:08:36.250 ⇒ 00:08:37.770 Pranav Narahari: And, you know…
61 00:08:37.770 ⇒ 00:08:41.029 Awaish Kumar: I… my point is that when we were discussing that, like.
62 00:08:41.270 ⇒ 00:08:43.770 Awaish Kumar: That was what I was expecting, that…
63 00:08:44.360 ⇒ 00:08:47.360 Awaish Kumar: So, from that conversation, the conclusion was, kind of.
64 00:08:47.480 ⇒ 00:08:57.390 Awaish Kumar: okay, we will use data at runtime. So I didn’t, like, made any objection, because, kind of, like, you guys are expert in AI, but…
65 00:08:57.500 ⇒ 00:09:07.489 Awaish Kumar: like, I was concerned about the documents, right? Because you mentioned that the scope of this is all the documents on Drive, all these,
66 00:09:10.110 ⇒ 00:09:17.779 Awaish Kumar: data that might live in G drive, emails, and then Slack, and then all those workspaces that they use.
67 00:09:17.990 ⇒ 00:09:21.449 Awaish Kumar: So the… I was… I was concerned that, like.
68 00:09:22.030 ⇒ 00:09:27.079 Awaish Kumar: At the runtime, if agent has to read all these documents to give you an answer.
69 00:09:27.960 ⇒ 00:09:28.690 Pranav Narahari: Okay.
70 00:09:28.820 ⇒ 00:09:29.920 Awaish Kumar: Reliable, right?
71 00:09:30.380 ⇒ 00:09:38.409 Pranav Narahari: So, I mean, okay, that’s good. Like, that we’re kind of addressing concerns now. This is, I think.
72 00:09:38.760 ⇒ 00:09:52.579 Pranav Narahari: me, you, and Sam, we need to discuss tomorrow, because I do kind of see your point where, first, AI needs to look at this, right? Sam needs to give you, hey, I want to take… I think the best approach for this.
73 00:09:52.580 ⇒ 00:10:08.109 Pranav Narahari: is RAG. I think the best approach for this is, you know, some other MCP, whatever it is, right? And then he can come to you with, like, okay, how do we land this data? What is the proper way to, you know, store these embeddings? Is that kind of what you’re saying?
74 00:10:09.870 ⇒ 00:10:16.329 Awaish Kumar: Yeah, number… number one thing is that, right? As an… Yeah. I need help from…
75 00:10:16.510 ⇒ 00:10:18.550 Awaish Kumar: for example, Sam, to come up with
76 00:10:18.700 ⇒ 00:10:36.000 Awaish Kumar: the solution, like, what actually my LLM, how it’s going to work. Then I’m kind of, okay, how I can bring data for you. That’s one part of our discussion that we can discuss tomorrow with Sam. I’m just brainstorming this project as a whole. I’m more thinking about now, like.
77 00:10:36.570 ⇒ 00:10:41.070 Awaish Kumar: if, like, cutser, if it has to read thousand documents.
78 00:10:42.180 ⇒ 00:10:45.640 Awaish Kumar: Like, it will, like, take maybe a minute to read it.
79 00:10:45.820 ⇒ 00:10:52.619 Awaish Kumar: Do you think the client will wait for that, or what is the expectation from the client?
80 00:10:53.790 ⇒ 00:11:01.040 Pranav Narahari: Yeah, so, like, for a chatbot, it can’t probably have… I mean, it can have a minute, right, if it’s doing deep analysis.
81 00:11:02.350 ⇒ 00:11:04.870 Pranav Narahari: It’s honestly fine. Yeah.
82 00:11:06.130 ⇒ 00:11:06.710 Awaish Kumar: Okay, so.
83 00:11:06.710 ⇒ 00:11:07.105 Pranav Narahari: So…
84 00:11:08.210 ⇒ 00:11:08.900 Awaish Kumar: Yeah, I’m…
85 00:11:08.900 ⇒ 00:11:10.329 Pranav Narahari: You just wait for a minute, right?
86 00:11:10.840 ⇒ 00:11:16.949 Awaish Kumar: I’m just trying to learn myself here. So basically, when I’m curing cursor, if it has to read, like.
87 00:11:17.200 ⇒ 00:11:19.190 Awaish Kumar: 1,000 documents, do you think it…
88 00:11:19.690 ⇒ 00:11:24.540 Awaish Kumar: like, how it reads, basically. Can you just brief me about that? A little bit?
89 00:11:24.640 ⇒ 00:11:30.340 Awaish Kumar: Like, how these agents actually read all those documents, if we don’t involve, like, the Ag or anything?
90 00:11:34.190 ⇒ 00:11:49.900 Pranav Narahari: So basically, with a… let’s say for CLI, right, for example, there’s probably certain metadata associated with each one of those documents that is very low in data. So it can… for those thousand documents, it can maybe take, let’s just say, a thousand
91 00:11:51.900 ⇒ 00:12:03.629 Pranav Narahari: 500-character strings, and then based on the query, it can then map that to the right document. And so maybe, like, that’s a very basic example.
92 00:12:03.630 ⇒ 00:12:06.090 Awaish Kumar: So, that metadata, where it comes from, right?
93 00:12:07.800 ⇒ 00:12:10.300 Pranav Narahari: like, this CLI,
94 00:12:10.690 ⇒ 00:12:21.590 Pranav Narahari: will potentially, like, just create a command based on the metadata that it knows it has. So with, like, the GW CLI, right, we’re not building that CLI. The CLI already exists.
95 00:12:21.590 ⇒ 00:12:25.980 Awaish Kumar: Yeah, okay, so we will not be even giving the metadata, it will just…
96 00:12:26.180 ⇒ 00:12:29.490 Awaish Kumar: use the CLI to get the metadata of the documents.
97 00:12:29.660 ⇒ 00:12:33.960 Awaish Kumar: And then shortlist based on carry, and then read those documents, right?
98 00:12:35.340 ⇒ 00:12:37.049 Pranav Narahari: Yeah, that… I mean, that’s one…
99 00:12:38.040 ⇒ 00:12:55.380 Pranav Narahari: that’s how the CLI works, right? How a command center should work for Eden is potentially we need to create our own knowledge base, our own… you can call it metadata for, like, our example here. So that’s where Sam kind of comes in and is like, okay, how do we then…
100 00:12:55.500 ⇒ 00:13:00.530 Pranav Narahari: Architect that? Like, what is the right schema for this, for this metadata?
101 00:13:00.770 ⇒ 00:13:05.810 Pranav Narahari: And now the metadata is not necessarily gonna be just for Google
102 00:13:05.970 ⇒ 00:13:14.889 Pranav Narahari: you know, Google Workspace documents, or, you know, calendar events, or whatever, it can actually be for just the Eden organization. So…
103 00:13:14.960 ⇒ 00:13:26.399 Pranav Narahari: One thing that I realized today is that the AI does a good job of summarizing just conversation, but it has no clue the topics in which it’s summarizing. And some… well.
104 00:13:26.480 ⇒ 00:13:33.600 Pranav Narahari: And they’re pretty smart, too. Sometimes it actually is like, okay, based on the conversations that are being had, this is likely the project that you’re working on.
105 00:13:33.700 ⇒ 00:13:36.630 Pranav Narahari: But there’s no source of truth to fall back on.
106 00:13:36.770 ⇒ 00:13:43.310 Pranav Narahari: So, let me give you an example. With, with Brainforge right now, we have these project roadmaps, right?
107 00:13:43.500 ⇒ 00:13:47.629 Pranav Narahari: They are a good source of truth of what we are working on day to day.
108 00:13:47.800 ⇒ 00:13:54.070 Pranav Narahari: So, like, me and you discussing this right now is gonna fall within the tree where the…
109 00:13:54.370 ⇒ 00:13:59.689 Pranav Narahari: you know, the parent node is that… is the pro… is the AI… Eden AI project.
110 00:14:00.140 ⇒ 00:14:03.040 Pranav Narahari: So…
111 00:14:03.680 ⇒ 00:14:14.640 Pranav Narahari: you can kind of think of every organization as gonna have some type of structure like that, you know? It’s probably gonna be, you know, like, quarterly memos that are sent out, maybe ad hoc.
112 00:14:14.880 ⇒ 00:14:30.450 Pranav Narahari: ELT, emails sent out that are like, hey, these are our initiatives for the next few months, not just for the next couple days. Having… and… but those are also going to be very finite, right? It’s going to be a much smaller document set.
113 00:14:30.610 ⇒ 00:14:36.440 Pranav Narahari: So… adding that as context to the AI, so then we can then…
114 00:14:36.570 ⇒ 00:14:40.290 Pranav Narahari: You know, not just search every single person’s…
115 00:14:40.520 ⇒ 00:14:50.719 Pranav Narahari: Slack message. We can be more focused in terms of whose Slack messages are we looking at? Whose Google Drive information? Whose Google Drive activity should we look at?
116 00:14:51.070 ⇒ 00:14:56.139 Pranav Narahari: that… that’s the goal here. I think that’s where…
117 00:14:56.560 ⇒ 00:15:03.810 Awaish Kumar: Yeah, I’m more… I’m confused at that point, like, how it defines, like, If, if you… So, we are…
118 00:15:04.160 ⇒ 00:15:08.809 Awaish Kumar: Like, okay, I’ll… let me give you an example, and then you tell me where we…
119 00:15:08.920 ⇒ 00:15:13.280 Awaish Kumar: are… we are trying to beam, right? So, we have, right now, we have a cursor.
120 00:15:13.510 ⇒ 00:15:21.109 Awaish Kumar: So, I’m kind of an end user. I’m giving an input to Cursor. Cursor uses maybe LLM in the backend.
121 00:15:21.310 ⇒ 00:15:29.659 Awaish Kumar: And it returns the responses. So with the new commander project, we are trying to be at the layer where the customer lives, right? No?
122 00:15:32.210 ⇒ 00:15:35.509 Pranav Narahari: Yeah, we could say that. Yeah, we’re trying to be a cursor.
123 00:15:36.530 ⇒ 00:15:37.210 Awaish Kumar: Okay.
124 00:15:37.780 ⇒ 00:15:42.609 Awaish Kumar: So, basically, the end user will give a query, and they… And in…
125 00:15:43.120 ⇒ 00:15:46.329 Awaish Kumar: And… and similar to Khashar, we will…
126 00:15:46.590 ⇒ 00:15:53.899 Awaish Kumar: be kind of defining these rules, like, so if it is about Eden AI, whose message it should read.
127 00:15:54.150 ⇒ 00:15:58.799 Awaish Kumar: this is something done by us, not by LLM by default.
128 00:16:00.650 ⇒ 00:16:02.699 Pranav Narahari: Whose messages should it read?
129 00:16:03.030 ⇒ 00:16:08.419 Pranav Narahari: We are going to give it structure in whose it looks at.
130 00:16:09.400 ⇒ 00:16:18.220 Pranav Narahari: Like, in that structure… that structure is gonna be, like, given by us. We can’t just be like, hey, if keyword this, then go look at this person.
131 00:16:18.260 ⇒ 00:16:32.470 Pranav Narahari: That’s not going to be scalable and maintainable. However, that’s where, like, kind of getting an organization structure of Eden comes into play. That’s where getting a… maybe a role description for each employee comes into play.
132 00:16:32.470 ⇒ 00:16:41.469 Pranav Narahari: Because those things an AI can do… can have very good judgment about, okay, based on this information, who is going to be associated with a certain project?
133 00:16:42.830 ⇒ 00:16:43.520 Awaish Kumar: Okay.
134 00:16:43.990 ⇒ 00:16:45.169 Pranav Narahari: Does that make sense?
135 00:16:45.890 ⇒ 00:16:46.850 Awaish Kumar: Yes, yes.
136 00:16:46.850 ⇒ 00:16:47.970 Pranav Narahari: Okay, cool.
137 00:16:47.970 ⇒ 00:16:50.859 Awaish Kumar: This happens at a cursor level, right? So I understand, so…
138 00:16:51.090 ⇒ 00:16:54.170 Awaish Kumar: At a command center level, we will be giving the…
139 00:16:55.470 ⇒ 00:17:04.930 Awaish Kumar: Okay, we will be giving… for example, a user carries about Eden AI project, but then there is a document that defines… on Eden AI project, there are, like, these
140 00:17:05.170 ⇒ 00:17:08.550 Awaish Kumar: 10 people involved, 5 of them belong to Brainford.
141 00:17:08.680 ⇒ 00:17:10.790 Awaish Kumar: 5 of them belong to Eden.
142 00:17:11.220 ⇒ 00:17:13.920 Awaish Kumar: Who is the CEO, CFO, whatever, and then…
143 00:17:15.300 ⇒ 00:17:20.119 Awaish Kumar: then LLM will determine, okay, these are the 10 people I should
144 00:17:20.250 ⇒ 00:17:25.500 Awaish Kumar: Messages only for these 10 people, if the cure is regarding this project, right?
145 00:17:26.380 ⇒ 00:17:31.300 Pranav Narahari: Yeah, it’s basically… I mean, I don’t know how that really… Helps with, like…
146 00:17:32.050 ⇒ 00:17:37.269 Pranav Narahari: kind of architecting, like, this database, like, the data part of things,
147 00:17:37.510 ⇒ 00:17:48.979 Pranav Narahari: Maybe it does, but yeah, I mean, more or less, that’s what it does. And currently, how the app is functioning is that it just doesn’t have that context, and it’s still able to give good insights.
148 00:17:49.060 ⇒ 00:18:07.820 Pranav Narahari: You know, because… but it just doesn’t have the structure to know, like, okay, these are the projects that are going on. Sometimes it is able to synthesize that information on its own, because there’s enough context within Slack or within email to get a good understanding of what the project is about and who’s involved in the project.
149 00:18:08.050 ⇒ 00:18:17.949 Pranav Narahari: However, we shouldn’t… we… we… the whole idea of having a knowledge base is that this is the source of truth. The LLM doesn’t define what the source of truth is.
150 00:18:18.790 ⇒ 00:18:19.400 Awaish Kumar: Okay.
151 00:18:21.260 ⇒ 00:18:36.540 Pranav Narahari: I also, I have to jump in, like, 2 minutes, but yeah, if you have one more question, yeah, we can go over it. And then also, there is the… I think the transcript of the meeting that I just had with Danny is going to be super helpful for you as well. So if you want to.
152 00:18:36.540 ⇒ 00:18:39.850 Awaish Kumar: The hospital center is a… App is already up.
153 00:18:40.800 ⇒ 00:18:43.270 Pranav Narahari: Yes, yes, yes, let me, let me send you a link.
154 00:18:43.620 ⇒ 00:18:44.970 Awaish Kumar: Okay, sure, thank you.
155 00:18:45.510 ⇒ 00:18:56.569 Pranav Narahari: Yeah, no problem. And I can still, I just have to hop off the call for a little bit. We can even… if you’re available a little bit later, we can hop back onto a call. I’ll just be, like, joining from my phone.
156 00:18:57.410 ⇒ 00:19:06.039 Awaish Kumar: Yeah, no worries, I will just look at the app and the… and your meeting with Danny, and yeah, I will try to understand what’s going on.
157 00:19:06.700 ⇒ 00:19:16.049 Pranav Narahari: Okay, cool. And I will also just schedule a call for the three of us tomorrow, just to make sure we can, you know… I’ll try to schedule it as early as possible tomorrow.
158 00:19:16.480 ⇒ 00:19:17.710 Awaish Kumar: Okay, thank you.
159 00:19:18.120 ⇒ 00:19:22.380 Pranav Narahari: Perfect, yeah. Thanks, Loish. Thanks for, thanks for, messaging me about this. Talk to you later.
160 00:19:22.750 ⇒ 00:19:23.719 Awaish Kumar: We’ll do what?
161 00:19:24.160 ⇒ 00:19:24.690 Pranav Narahari: Chill.