Meeting Title: Sync on implementation plan for eden Date: 2026-03-25 Meeting participants: Pranav, Samuel Roberts, Awaish Kumar
WEBVTT
1 00:00:56.430 ⇒ 00:00:57.430 Samuel Roberts: Ayy.
2 00:00:58.200 ⇒ 00:00:59.390 Pranav: Hey, Sam. Morning.
3 00:01:00.500 ⇒ 00:01:01.240 Samuel Roberts: How are you?
4 00:01:01.750 ⇒ 00:01:02.569 Pranav: Pretty good.
5 00:01:03.850 ⇒ 00:01:10.279 Pranav: Yeah, so… I don’t know if anybody else is gonna join, I didn’t see any.
6 00:01:10.280 ⇒ 00:01:10.890 Samuel Roberts: Okay.
7 00:01:10.890 ⇒ 00:01:12.309 Pranav: Accept the meeting.
8 00:01:12.310 ⇒ 00:01:24.010 Samuel Roberts: Yeah, yeah, I’m not even sure if I did, to be honest. Sorry. Yeah, so I saw Robert’s message. I think that’s kind of… I wasn’t 100% sure, like, where to go with the second part.
9 00:01:24.640 ⇒ 00:01:25.240 Samuel Roberts: Partly…
10 00:01:25.240 ⇒ 00:01:26.189 Pranav: Yeah, so I think…
11 00:01:26.190 ⇒ 00:01:27.279 Samuel Roberts: Yeah, go ahead.
12 00:01:27.280 ⇒ 00:01:33.930 Pranav: What might be useful, I’m gonna read through it right now, if you can also just, like, read through that document about, like.
13 00:01:34.240 ⇒ 00:01:41.780 Pranav: So Robert is saying, like, yeah, he’s not going to come up with a road plan until we have the data, so…
14 00:01:42.450 ⇒ 00:01:46.340 Pranav: I guess I just need to kind of understand this document, like…
15 00:01:46.450 ⇒ 00:01:52.300 Pranav: I’m trying to read this to see, like, okay, what can Robert add if he needs to add anything right now?
16 00:01:54.300 ⇒ 00:02:00.429 Pranav: What we can say is, like, okay, once we have the data, then we generate this type of data analysis roadmap.
17 00:02:01.070 ⇒ 00:02:06.720 Pranav: Which we then will use to implement the…
18 00:02:07.050 ⇒ 00:02:10.369 Pranav: the AI analysis, and then create the report.
19 00:02:10.370 ⇒ 00:02:14.660 Samuel Roberts: Yeah, sorry, one second, just gotta find the right… here we go. Sorry, the wrong one open. Okay.
20 00:02:18.020 ⇒ 00:02:19.020 Pranav: So… Yeah.
21 00:02:19.960 ⇒ 00:02:20.970 Pranav: I think…
22 00:02:21.230 ⇒ 00:02:30.640 Pranav: what’s gonna be important for you here is this is all kind of like, data analysis and reporting and AI,
23 00:02:30.820 ⇒ 00:02:41.599 Pranav: projects, like that second project. So, kind of understanding, like, okay, with this type of data analysis roadmap, what are the gaps that need to be filled in with AI?
24 00:02:41.880 ⇒ 00:02:43.000 Pranav: So…
25 00:02:43.890 ⇒ 00:02:55.680 Pranav: there’s probably a little bit of a data gap as well, which I think we… you can ask Oish about to fill in on, or to get his tech… I think what would be best is probably you do a first pass of
26 00:02:55.900 ⇒ 00:03:01.640 Pranav: I think you have a pretty good… considering, like, how we did things with Lilo, like, understanding of how to…
27 00:03:02.160 ⇒ 00:03:20.539 Pranav: think about data as well. So, what are the data inputs that Robert needs to create this data analysis roadmap? What is then the output that we can then use, or the process that’s going to be created in this data analysis roadmap, that we can then use AI to generate the actual reports? So…
28 00:03:21.010 ⇒ 00:03:22.110 Pranav: kind of…
29 00:03:22.840 ⇒ 00:03:33.279 Pranav: figure out, and I’ll help you with this too, like, with this data analysis roadmap that Robert’s gonna make, how does that tie into the AI
30 00:03:34.340 ⇒ 00:03:35.430 Pranav: projects. So…
31 00:03:36.060 ⇒ 00:03:38.140 Samuel Roberts: Let me just… I have a couple questions. So, what…
32 00:03:38.240 ⇒ 00:03:41.389 Samuel Roberts: when I was looking through it, it talked about, like, metadata.
33 00:03:42.430 ⇒ 00:03:46.770 Samuel Roberts: Calendar… I’m just trying to see, like, what sorts of data we’re getting here.
34 00:03:47.390 ⇒ 00:03:49.950 Samuel Roberts: Gmail thread length of receipt.
35 00:03:50.890 ⇒ 00:03:52.710 Samuel Roberts: Recipient density…
36 00:03:57.080 ⇒ 00:04:02.979 Samuel Roberts: I think I just… I don’t have a great sense of, like, what data we’re really looking for, and, like, what… what is helpful for…
37 00:04:03.220 ⇒ 00:04:05.690 Samuel Roberts: What’s his name? Danny?
38 00:04:06.130 ⇒ 00:04:13.640 Samuel Roberts: In terms of, like, what… what will be extracted, and, like, what… you know, I… there’s just a lot of pieces moving around here that I’m not sure where…
39 00:04:14.140 ⇒ 00:04:16.720 Samuel Roberts: Like, what we will get out of this, like…
40 00:04:17.290 ⇒ 00:04:21.489 Samuel Roberts: Are we just looking more at metadata for a lot of this stuff? Comment resolution rate…
41 00:04:23.150 ⇒ 00:04:30.559 Pranav: So, yeah, all those different things are different ways for us to just see how projects at Eden are progressing.
42 00:04:30.850 ⇒ 00:04:31.410 Pranav: Right.
43 00:04:31.410 ⇒ 00:04:34.720 Samuel Roberts: The main goal, is just, like, how things are moving. Okay.
44 00:04:34.910 ⇒ 00:04:36.779 Pranav: How things are moving, yeah.
45 00:04:36.780 ⇒ 00:04:37.520 Samuel Roberts: Anyways…
46 00:04:37.980 ⇒ 00:04:38.839 Pranav: Hey, I wish.
47 00:04:39.080 ⇒ 00:04:44.390 Pranav: Yeah, so we wanted to see how things were moving, so seeing the…
48 00:04:44.730 ⇒ 00:04:53.809 Pranav: the amount… and there’s so much analysis that we can do just based off Slack messages as well, right? We can look at the exact messages that are being,
49 00:04:53.920 ⇒ 00:05:01.430 Pranav: that are being sent back and forth on a certain project. We can assess, is this looking like they’re putting out fires, or does it look like that they’re.
50 00:05:01.430 ⇒ 00:05:02.609 Samuel Roberts: Mmm, very little.
51 00:05:02.610 ⇒ 00:05:04.430 Pranav: Happening on Slack.
52 00:05:04.610 ⇒ 00:05:08.260 Pranav: So, I think that’s kind of where we need to put our, like…
53 00:05:08.830 ⇒ 00:05:11.640 Pranav: Analysis hat on to see, like, okay.
54 00:05:11.910 ⇒ 00:05:31.619 Pranav: what do we even notice here at Brainforge with… okay, like, and we can even use this as, like, a meta example, right? Creating this project plan, like, okay, we’re putting comments on the project plan, we’re also talking about it via Slack, we’re creating Google Calendar events to discuss this one project.
55 00:05:31.720 ⇒ 00:05:32.580 Pranav: Right?
56 00:05:33.070 ⇒ 00:05:35.440 Pranav: This applies across
57 00:05:35.550 ⇒ 00:05:49.629 Pranav: every other project, too, here at Brainforge, but then also at every other company, too. And so what they’re trying to do at Eden is, like, how does the COO get a high-level understanding of the health going on for each project?
58 00:05:51.210 ⇒ 00:05:51.980 Awaish Kumar: I.
59 00:05:51.980 ⇒ 00:05:56.430 Pranav: Something unhealthy is probably like, okay, deadlines keep on getting pushed, if, like, there’s a lot of Slack community…
60 00:05:57.470 ⇒ 00:05:58.850 Pranav: Figure out.
61 00:05:58.850 ⇒ 00:06:00.209 Samuel Roberts: Is that better? Absolutely.
62 00:06:00.430 ⇒ 00:06:03.020 Pranav: Does that kind of paint the picture a little bit better?
63 00:06:03.190 ⇒ 00:06:11.359 Samuel Roberts: Yeah, yeah, because it was, like, the wording just, like, themes across business, I was just, like, I didn’t fully understand, and I… maybe it’s just because I wasn’t…
64 00:06:11.530 ⇒ 00:06:16.249 Samuel Roberts: you know, in some of those meetings and stuff, but I think that makes more sense now. Okay. Okay. Yeah.
65 00:06:16.250 ⇒ 00:06:25.750 Pranav: And to get further, kind of, like, context, too, like, take a look and, like, just… I think Cursor’s gonna be your best friend here, just, like, on that call that,
66 00:06:26.020 ⇒ 00:06:31.839 Pranav: Robert and I have had with Danny, so there’s a couple of calls,
67 00:06:31.940 ⇒ 00:06:44.370 Pranav: to kind of just get a better understanding of, like, okay, what is the data that they’re looking for? Where does this pain come from? I think some of that information is going to be in that project plan, some of it’s going to be in the SOW,
68 00:06:44.720 ⇒ 00:06:55.090 Pranav: but also just, like, kind of getting maybe, like… you can… also, from just the tone and just, like, the specific wording that Danny uses, you might get some information from that as well.
69 00:06:57.170 ⇒ 00:07:02.699 Awaish Kumar: I, you know… Runav, can we discuss about the… how you are looking…
70 00:07:03.090 ⇒ 00:07:07.510 Awaish Kumar: to get the data. I just spoke with Utam before this call.
71 00:07:07.650 ⇒ 00:07:10.349 Awaish Kumar: Like, he told me that, like.
72 00:07:11.480 ⇒ 00:07:15.960 Awaish Kumar: Craig, there are two pieces of this project. One is data.
73 00:07:17.270 ⇒ 00:07:20.009 Awaish Kumar: Extraction, and building kind of a warehouse.
74 00:07:20.110 ⇒ 00:07:21.720 Awaish Kumar: for the AI to use.
75 00:07:22.250 ⇒ 00:07:25.590 Awaish Kumar: And for that, I need an answer from you.
76 00:07:25.800 ⇒ 00:07:28.269 Awaish Kumar: Where do you want this data to be?
77 00:07:28.550 ⇒ 00:07:34.389 Awaish Kumar: In what format? Like, do you want files? Do you want tables? What… in what format do you need?
78 00:07:34.630 ⇒ 00:07:36.370 Awaish Kumar: Right?
79 00:07:36.370 ⇒ 00:07:37.390 Pranav: Okay, yeah.
80 00:07:39.710 ⇒ 00:07:40.280 Awaish Kumar: Don’t.
81 00:07:40.280 ⇒ 00:07:46.450 Pranav: What I initially thought about, right, was, like, when I was writing out this project plan, like, I was kind of just writing it end-to-end.
82 00:07:47.420 ⇒ 00:07:58.520 Pranav: warehouse, okay, made sense to me, that’s what I’ve done in the past. Then Uten comes to me and is like, do you need a data warehouse for this? Why can’t we just use the CLI and query the data ad hoc when you’re wanting to generate the report?
83 00:07:58.640 ⇒ 00:08:10.829 Pranav: And I was like, yeah, maybe we can do that. I don’t understand… I couldn’t give a clear answer based on the complexity of what we’re trying to build here. So that’s why, like, Oish, in that message, I was trying to describe the exact.
84 00:08:10.830 ⇒ 00:08:11.370 Samuel Roberts: Hmm…
85 00:08:11.370 ⇒ 00:08:14.540 Pranav: product deliverables that we’re looking for. This is what…
86 00:08:14.800 ⇒ 00:08:22.130 Awaish Kumar: I get it, but, we have to, like, split this, right, into chunks to actually
87 00:08:22.390 ⇒ 00:08:26.029 Awaish Kumar: Come up with… come up with some technical solution, right?
88 00:08:27.290 ⇒ 00:08:34.250 Awaish Kumar: From deliverable, we have to go and actually, like, divide it into milestones, right?
89 00:08:34.390 ⇒ 00:08:38.210 Awaish Kumar: For example, it says, Calendar data, right?
90 00:08:38.620 ⇒ 00:08:40.700 Awaish Kumar: It mentions Gmail data.
91 00:08:40.820 ⇒ 00:08:43.840 Awaish Kumar: Slack data and the files from Drive.
92 00:08:43.940 ⇒ 00:08:56.009 Awaish Kumar: So, these are the four… I see these are the four sources. Now, what I need from you is, like, for example, okay, if you confirm, these are only sources from where you need the data.
93 00:08:56.510 ⇒ 00:09:02.470 Awaish Kumar: And then where it should be, like, it could be, In the…
94 00:09:02.720 ⇒ 00:09:09.339 Awaish Kumar: warehouse, it could be in a S3 file, like, for… like, previously we had these, like,
95 00:09:09.490 ⇒ 00:09:26.180 Awaish Kumar: conversation as .txt files in Superbase, right? So, it all depends how you are using it right now. Maybe you have done it for ABC, or if you have done it for an internal platform, whatever, like, you think works for AI piece, I don’t know what… how Gemini…
96 00:09:26.350 ⇒ 00:09:27.060 Awaish Kumar: Handle that?
97 00:09:27.060 ⇒ 00:09:34.560 Pranav: Yeah, I think that’s where Sam can help you a little bit more, too. So, from… from my perspective, right, like, I kind of…
98 00:09:34.760 ⇒ 00:09:35.780 Pranav: I can…
99 00:09:35.880 ⇒ 00:09:47.989 Pranav: I kind of gave my input in terms of, like, what I think would be necessary for this project, but I don’t feel super confident in that. I think that’s where Utam is like, okay, get Awash and get Sam to discuss, like…
100 00:09:47.990 ⇒ 00:09:48.430 Samuel Roberts: Yeah, yeah.
101 00:09:48.430 ⇒ 00:10:05.260 Pranav: where AI fits in with the data. So, like, for me to give a… do I think this can exist in Subabase and as TXT files, does this exist as tables in a data warehouse? That’s kind of for y’all to have, like, the creative freedom to, like, build the system. What I am trying to…
102 00:10:05.260 ⇒ 00:10:10.840 Pranav: paint out for you guys, and let me know if, like, you need more context for this, is just, like, what the product needs to do.
103 00:10:11.450 ⇒ 00:10:14.290 Awaish Kumar: Yeah, so, from 2 to…
104 00:10:14.470 ⇒ 00:10:21.789 Awaish Kumar: to help with data piece, these are the things that I need. Like, if Sam can answer that, like, I’m…
105 00:10:21.930 ⇒ 00:10:24.889 Awaish Kumar: Like, I just need the… need to, need to know.
106 00:10:25.090 ⇒ 00:10:28.439 Awaish Kumar: that how AI is going to use that, that’s all, like…
107 00:10:28.440 ⇒ 00:10:38.809 Pranav: Okay, yeah, I think that’s… that’s exactly why I wanted to put you guys together. So, yeah, Sam, you’ll probably, like, Awash will probably, like, paint, like, a solution for, like, how we could store, like.
108 00:10:39.020 ⇒ 00:10:40.810 Pranav: In… if we…
109 00:10:41.030 ⇒ 00:10:48.289 Pranav: how we can store the data, I guess. The simplest solution is just using the CLI to query the data.
110 00:10:48.540 ⇒ 00:10:49.050 Pranav: Right.
111 00:10:49.050 ⇒ 00:10:49.950 Awaish Kumar: But…
112 00:10:49.950 ⇒ 00:10:50.500 Pranav: if…
113 00:10:50.820 ⇒ 00:11:04.099 Awaish Kumar: That is a different, like, the… what Utam is… what Utam is saying is that at the query time, you use CLI to read those files, but if there are thousands of files, like.
114 00:11:04.100 ⇒ 00:11:05.599 Samuel Roberts: Yeah, I don’t know if that’s a good… I think.
115 00:11:06.000 ⇒ 00:11:11.100 Awaish Kumar: You can… you will be… read that all there, and try to figure out an answer.
116 00:11:11.280 ⇒ 00:11:15.069 Awaish Kumar: Using… like, in the real time, you… I don’t think you can do that, right?
117 00:11:15.220 ⇒ 00:11:33.370 Pranav: Well, so there’s two aspects to this, right? Like, we are not trying to be able to generate the… do the deem detection in real time. That’s not part of the project. The… the query part of it, like, in a chat interface, is just if you’re trying to get a specific piece of information.
118 00:11:33.590 ⇒ 00:11:41.409 Awaish Kumar: Yeah, for example, Pranob, if I… I’m a city… like Danny, for example, if I say, okay, I need information regarding
119 00:11:42.760 ⇒ 00:11:48.229 Awaish Kumar: XYZ, like, Eden OS project that’s going on. Okay, I want to see the progress of that project.
120 00:11:48.560 ⇒ 00:11:49.890 Awaish Kumar: What’s happening?
121 00:11:50.250 ⇒ 00:11:56.490 Awaish Kumar: So… how the AI will work. It needs to go to those documents and read it, right?
122 00:11:56.490 ⇒ 00:11:57.090 Pranav: Right.
123 00:11:57.890 ⇒ 00:12:00.689 Awaish Kumar: So what document it should read? It doesn’t know, like.
124 00:12:02.390 ⇒ 00:12:12.779 Awaish Kumar: like, by default, right? It has to read everything to come up with that, or you use some kind of rag or something like that to come up with that filtering, right?
125 00:12:13.610 ⇒ 00:12:20.099 Awaish Kumar: And then you can say, okay, this… these are the… out of thousand documents, these are the four ones where I got this answer for you.
126 00:12:21.820 ⇒ 00:12:28.030 Pranav: So, I guess, Sam, for using the CLI, I assume that there would be some type of… like…
127 00:12:28.140 ⇒ 00:12:37.159 Pranav: query generation that happens based on the input of the user, right? So if they’re asking about a certain specific project, then
128 00:12:37.520 ⇒ 00:12:42.650 Pranav: the CLI will then… Extract only the relevant files.
129 00:12:43.300 ⇒ 00:12:53.830 Pranav: And we can add some cap to that as well, if we’re trying to support something in real time, right? So, like, we’ll make sure it doesn’t go across 50 files of context, something like that.
130 00:12:54.300 ⇒ 00:12:59.429 Pranav: And so that is, like… That is, like, the use case for chat.
131 00:12:59.500 ⇒ 00:13:17.309 Pranav: But then for theme detection, that doesn’t need to happen in real time. That can be a process that… you can use a CLI, potentially, you can tell me the limitations of the CLI, to extract maybe a thousand files, and then that… all those files get processed, extracts the data, and then…
132 00:13:17.390 ⇒ 00:13:28.129 Pranav: then we can find the clustering, do the theme detection that way. But that doesn’t need to happen real time, so we don’t need to, like… because that may take, like, 5, 10, maybe an hour.
133 00:13:29.240 ⇒ 00:13:34.819 Awaish Kumar: No, but I don’t think that is a feasible solution. Every time someone carries, you go back to reading all those
134 00:13:35.240 ⇒ 00:13:39.879 Awaish Kumar: 1,000 files, creating… Right for this.
135 00:13:39.880 ⇒ 00:13:43.089 Samuel Roberts: for the theme analysis, I think we definitely want to store
136 00:13:43.410 ⇒ 00:13:49.739 Samuel Roberts: Because especially we want, like, historical stuff, too, right? I don’t know if the CLI would have everything that, like, has changed over time, even.
137 00:13:50.090 ⇒ 00:13:51.230 Samuel Roberts: Yeah.
138 00:13:51.410 ⇒ 00:13:58.739 Pranav: And so maybe it’s a combination of the both, right? Like, maybe use a CLI for just chat, maybe use a data warehouse. I think…
139 00:13:58.810 ⇒ 00:14:12.580 Pranav: what I’ve tried to map out here is the timeline, right, and the milestones. And the milestones, too, you can let me know, like, if we cannot hit a certain milestone without severely, like, veering off the path of, like, the main goal.
140 00:14:13.280 ⇒ 00:14:15.530 Pranav: And so…
141 00:14:15.780 ⇒ 00:14:16.350 Samuel Roberts: I see.
142 00:14:17.930 ⇒ 00:14:23.649 Pranav: Yeah, I think this is, like, a good conversation. I think, Sam, you have all the kind of AI context and, like.
143 00:14:23.840 ⇒ 00:14:27.629 Pranav: deep understanding of, like, what AI can do, and fit into the picture with
144 00:14:27.750 ⇒ 00:14:37.940 Pranav: And what are the data requirements for the AI to work? So, yeah, I think you too, like, having that conversation, and then… yeah, if I can help in any way, I will help, but…
145 00:14:38.510 ⇒ 00:14:42.619 Pranav: I think this is where the conversation needs to be had, and then a system design is gonna fall out of that.
146 00:14:43.920 ⇒ 00:14:50.260 Awaish Kumar: So, in my past experiences with the… with a little bit of idol of data platform, when… while it was…
147 00:14:50.880 ⇒ 00:14:51.930 Awaish Kumar: being built.
148 00:14:52.140 ⇒ 00:14:54.599 Awaish Kumar: So, what we did, basically, was…
149 00:14:56.220 ⇒ 00:15:01.639 Awaish Kumar: like, the AI team, like, KZ and, I don’t remember the name, the one.
150 00:15:02.440 ⇒ 00:15:10.569 Awaish Kumar: Like, dude… Basically, I bring that data to Superbase. What they did, basically, read those files.
151 00:15:10.800 ⇒ 00:15:16.329 Awaish Kumar: and created some, encodings, like RARG, RAG-based, or whatever.
152 00:15:16.640 ⇒ 00:15:20.560 Awaish Kumar: And based on that, they stored those in the tables.
153 00:15:20.830 ⇒ 00:15:26.830 Awaish Kumar: And when a user basically carries, they apply the same
154 00:15:27.170 ⇒ 00:15:38.160 Awaish Kumar: encodings, and then search using vector search or whatever, and come up with answers. That’s why, what, like, it has been happening before you came.
155 00:15:38.700 ⇒ 00:15:41.870 Awaish Kumar: I’m not sure if we are going to follow same process, because
156 00:15:42.750 ⇒ 00:15:46.309 Awaish Kumar: I’m… I’m not sure how… like, we do…
157 00:15:46.820 ⇒ 00:15:51.980 Awaish Kumar: that with Cursor, like, right now, we… we ask Cursor to directly connect with the…
158 00:15:52.080 ⇒ 00:15:56.940 Awaish Kumar: for example, all the files we have in our GitHub repo, It can read…
159 00:15:58.490 ⇒ 00:16:11.650 Awaish Kumar: those files, and if you are trying to do that, if you want to just use CLI, I don’t think you read them data in DataHelp, because with the CLI, if you have access to Google Drive, you can all… you can read all those files.
160 00:16:11.830 ⇒ 00:16:15.680 Awaish Kumar: Then we don’t need any data movement in that case.
161 00:16:19.440 ⇒ 00:16:20.719 Samuel Roberts: For the duplicate.
162 00:16:21.270 ⇒ 00:16:37.080 Awaish Kumar: Yeah, you don’t, like, I don’t know, like, what you are thinking, like, about AI piece. If you want to use lagger stuff, or you want to use the culture way that we are doing right now, like, right now, we just say, okay, there are a few files, 10, like, in my folder, just read all those.
163 00:16:37.390 ⇒ 00:16:39.209 Awaish Kumar: You can apply a similar approach.
164 00:16:39.730 ⇒ 00:16:44.060 Samuel Roberts: Right. Yeah, I haven’t used the CLI a ton yet, so I don’t know, like…
165 00:16:45.540 ⇒ 00:16:49.070 Samuel Roberts: how deep it can go, I guess, if that makes sense?
166 00:16:49.070 ⇒ 00:16:55.549 Awaish Kumar: But the, like, what Utam was just saying, that you… like, you have done kind of similar exercise for ABC, or some…
167 00:16:56.160 ⇒ 00:17:00.549 Awaish Kumar: And we can… You used that approach? I don’t know what you did for ABC.
168 00:17:01.490 ⇒ 00:17:04.540 Samuel Roberts: And I’m… I don’t know… What that’s referring to.
169 00:17:05.000 ⇒ 00:17:06.490 Awaish Kumar: Yeah, but… With the CLI?
170 00:17:07.270 ⇒ 00:17:08.800 Awaish Kumar: No, no, like…
171 00:17:09.290 ⇒ 00:17:10.720 Samuel Roberts: Oh, the rag approach?
172 00:17:12.839 ⇒ 00:17:18.139 Awaish Kumar: I don’t know, he mentioned that you have done similar, like, chat interface for ABC, you know?
173 00:17:18.140 ⇒ 00:17:20.490 Pranav: Yeah, so for, I think, Sam.
174 00:17:20.490 ⇒ 00:17:21.140 Samuel Roberts: Oh, I see.
175 00:17:21.140 ⇒ 00:17:23.440 Pranav: Talk about the rag approach on the central docks, right?
176 00:17:23.440 ⇒ 00:17:24.290 Samuel Roberts: Yeah, that makes sense.
177 00:17:24.290 ⇒ 00:17:32.580 Pranav: docs. Yeah, so I mean, one thing that I’m thinking about, right, is Gemini currently has custom integrations with
178 00:17:32.800 ⇒ 00:17:37.569 Pranav: Cool, like, you literally just, like, flip on a switch, you can get the context of your calendar, of your drive.
179 00:17:37.570 ⇒ 00:17:38.380 Samuel Roberts: Right.
180 00:17:38.380 ⇒ 00:17:38.890 Pranav: Jim and I have.
181 00:17:38.890 ⇒ 00:17:40.030 Samuel Roberts: that, yeah.
182 00:17:40.030 ⇒ 00:17:42.570 Pranav: of your mail, right? And so…
183 00:17:43.440 ⇒ 00:17:57.820 Pranav: what we can… and Sam, maybe what you can look into is, like, how are they doing that, right? Like, there’s probably some docs online, or just, like… this is probably a common problem of just, like, when you have, like, a massive data store like that, and you create, like, you have that integration.
184 00:17:58.010 ⇒ 00:18:02.079 Pranav: Do you embed all that information? Do you…
185 00:18:02.480 ⇒ 00:18:06.239 Pranav: how exactly are you pulling in that context for the AI?
186 00:18:07.290 ⇒ 00:18:21.030 Pranav: I think that… if we figure out what solution works best there, then… okay, then, in that situation, we’re likely… maybe we run the embeddings in real time? I don’t know.
187 00:18:21.350 ⇒ 00:18:29.909 Pranav: And then that way we get away from having to have a data warehouse or an embeddings database.
188 00:18:30.210 ⇒ 00:18:35.170 Pranav: So that’s an option, but I guess that’s kind of how we have to think about it, I guess. Like, really…
189 00:18:35.620 ⇒ 00:18:44.119 Pranav: think about, like, the individual steps of, okay, we pull in these hundred files as context for a query about, okay, we’re asking about Eden OS, right?
190 00:18:44.370 ⇒ 00:18:55.639 Pranav: okay, all these files we get, you know, from different sources too, from Slack, from Google, from, from Mail, whatever.
191 00:18:56.130 ⇒ 00:18:59.719 Pranav: Then, how do we actually use that as… as context?
192 00:19:01.350 ⇒ 00:19:05.710 Awaish Kumar: I think one of you already got the access, so why not we just… just try?
193 00:19:05.860 ⇒ 00:19:06.680 Awaish Kumar: Something.
194 00:19:08.380 ⇒ 00:19:11.460 Pranav: Sure, you want me to hop into Gemini?
195 00:19:12.150 ⇒ 00:19:17.330 Awaish Kumar: No, like, maybe put those credentials in one pass, and
196 00:19:18.110 ⇒ 00:19:24.530 Awaish Kumar: Sam can go in and try out what is… when it integrates, what happens, how it works.
197 00:19:24.530 ⇒ 00:19:33.360 Samuel Roberts: Yeah, because that’s a good point, because… so, for ABC, we built our own, like, chat bot, right? This they’re talking about into the Gemini…
198 00:19:34.050 ⇒ 00:19:35.470 Samuel Roberts: Chat already, right?
199 00:19:35.470 ⇒ 00:19:46.930 Pranav: No, no, no, it doesn’t need to necessarily be like that. So, that is an option. If it’s not possible to do that, then that’s, like, a pride restriction, right? Like, we just… we have to have our own… Right, but that changes.
200 00:19:46.930 ⇒ 00:19:52.029 Samuel Roberts: that changes pretty dramatically how we have to go about dealing with the, like, Google Workspace stuff, for example.
201 00:19:55.010 ⇒ 00:19:59.130 Pranav: Yeah, we have to create custom… Integration is what you’re saying?
202 00:19:59.450 ⇒ 00:20:03.850 Samuel Roberts: Right, because if Gemini already has access to Google Workspace stuff… That…
203 00:20:03.850 ⇒ 00:20:06.820 Pranav: Well, it doesn’t have access to… like.
204 00:20:07.520 ⇒ 00:20:12.649 Pranav: it has access on a user basis, right? We’ll still have to get, like, the global information.
205 00:20:14.400 ⇒ 00:20:20.440 Samuel Roberts: Oh, I see what you’re saying. He has access to his own calendar, Gmail, drive. Yeah.
206 00:20:22.960 ⇒ 00:20:25.910 Pranav: No, I don’t know how much of that is configurable to create.
207 00:20:26.110 ⇒ 00:20:31.630 Pranav: To make it organization-wide, like, I don’t know if Gemini has that part built in.
208 00:20:32.350 ⇒ 00:20:37.970 Pranav: And then also to… to, like, what you were saying before about the CLI, like, how deep does it go? Does it just give you, like.
209 00:20:38.320 ⇒ 00:20:40.739 Pranav: Basic information, like…
210 00:20:42.350 ⇒ 00:20:44.620 Awaish Kumar: Yeah, like… Like, how can…
211 00:20:44.620 ⇒ 00:20:47.699 Pranav: going to, like, the comments of, like, a Google Doc, like…
212 00:20:47.960 ⇒ 00:20:48.840 Samuel Roberts: Right, right.
213 00:20:48.840 ⇒ 00:20:49.490 Pranav: Yeah.
214 00:20:49.980 ⇒ 00:20:57.150 Awaish Kumar: Okay. Can we split this into, like, maybe, Sam, you can do a bit of a spike, maybe, in our…
215 00:20:57.150 ⇒ 00:21:00.370 Samuel Roberts: Yeah, no, I need to dig in deeper to that, then. Okay, but if we’re… yeah, so…
216 00:21:00.550 ⇒ 00:21:08.749 Samuel Roberts: there’s two… the two kind of paths here. One is the Gemini already needs access, so maybe not Gemini then, just whatever chat agent we’re building.
217 00:21:10.110 ⇒ 00:21:18.669 Pranav: Yeah, you should think of, like, the requirements that are necessary by any chat interface, right? Like, whether we build it or them, and then we will see if Gemini can fit in, right?
218 00:21:18.670 ⇒ 00:21:20.680 Samuel Roberts: Got it, no, okay, that makes more sense. Okay, cool, cool.
219 00:21:22.150 ⇒ 00:21:28.509 Awaish Kumar: So, yeah, you can look… you can do a spike on how you are going to do that, and then…
220 00:21:28.660 ⇒ 00:21:38.289 Awaish Kumar: once I have an answer, what do you exactly need from me, I can go and then think about how can I bring data
221 00:21:38.410 ⇒ 00:21:40.120 Awaish Kumar: To wherever you need.
222 00:21:40.850 ⇒ 00:21:43.650 Awaish Kumar: Right now, it’s, like, we, we…
223 00:21:43.820 ⇒ 00:21:50.760 Awaish Kumar: it completely depends on, you know, how you… if you are going to directly connect, then I don’t have to move data, right? Right.
224 00:21:50.760 ⇒ 00:21:51.280 Samuel Roberts: Right, okay.
225 00:21:51.680 ⇒ 00:21:52.410 Awaish Kumar: If you…
226 00:21:52.410 ⇒ 00:21:52.790 Samuel Roberts: I mean.
227 00:21:52.960 ⇒ 00:21:53.969 Awaish Kumar: Somewhere we can…
228 00:21:53.970 ⇒ 00:21:57.930 Samuel Roberts: Slack doesn’t have a way to do that, though, does it? Because that’s the other piece here.
229 00:21:59.240 ⇒ 00:22:03.119 Awaish Kumar: Yeah, Sam, I think you have to look into when you get access.
230 00:22:03.120 ⇒ 00:22:06.980 Samuel Roberts: I’ll figure that out. I’ll figure out what we can get access to, which ways, and .
231 00:22:07.150 ⇒ 00:22:12.439 Awaish Kumar: the integrations, right? What are the built-in integrations? You can look at that, if that is,
232 00:22:12.660 ⇒ 00:22:19.159 Awaish Kumar: If you can find it there, that’s okay. If not, then we can build our own pipeline to bring in Slack data.
233 00:22:20.070 ⇒ 00:22:20.610 Samuel Roberts: Okay.
234 00:22:21.710 ⇒ 00:22:23.549 Samuel Roberts: Sounds good. I’ll do that.
235 00:22:25.220 ⇒ 00:22:28.920 Pranav: Awesome, guys. Yeah, and I’m around, too, like, if you need more information, like.
236 00:22:29.170 ⇒ 00:22:35.069 Pranav: where I can, like, help. I think… I think now, though, we have a little bit of better understanding of, like, okay.
237 00:22:35.070 ⇒ 00:22:36.250 Awaish Kumar: Is this.
238 00:22:36.250 ⇒ 00:22:37.680 Pranav: What information comes from who?
239 00:22:39.080 ⇒ 00:22:41.370 Awaish Kumar: Is this the project review,
240 00:22:41.570 ⇒ 00:22:43.610 Awaish Kumar: thing we are working on? I know, like, Uttra.
241 00:22:43.610 ⇒ 00:22:46.129 Samuel Roberts: Yeah, yeah, yeah, this is the Eden AI project.
242 00:22:46.560 ⇒ 00:22:53.470 Awaish Kumar: Yeah, but he put out a project review piece that we need to all sign off, right?
243 00:22:53.470 ⇒ 00:22:54.030 Samuel Roberts: Excellent.
244 00:22:54.270 ⇒ 00:22:54.790 Awaish Kumar: nonsense.
245 00:22:55.410 ⇒ 00:22:57.100 Pranav: Yeah, this is the project plan.
246 00:22:57.490 ⇒ 00:23:00.980 Awaish Kumar: Okay, okay, this is the plan that we have to sign off, right, at the end.
247 00:23:01.700 ⇒ 00:23:07.739 Pranav: Yep. Yeah, that doc that I shared with you guys yesterday, and I think, Awish, I tagged you in that as well, that’s, like…
248 00:23:08.200 ⇒ 00:23:13.460 Pranav: that’s the doc that we’re gonna finalize, hopefully by end of day today.
249 00:23:13.460 ⇒ 00:23:14.880 Awaish Kumar: Okay, this is the one, right?
250 00:23:14.880 ⇒ 00:23:18.220 Pranav: Yeah. This is… this is the one, yeah.
251 00:23:18.850 ⇒ 00:23:19.410 Awaish Kumar: Sure.
252 00:23:19.410 ⇒ 00:23:31.790 Pranav: There’s… basically, I fill out all of this stuff except for that technical approach section. That’s where you guys… I want to get y’all’s draft on, I’ll read it, I’ll make sure it aligns with, like, all the deliverables that we have in terms of milestones.
253 00:23:32.170 ⇒ 00:23:38.670 Pranav: And then… Yeah, we’ll… The goal is to kind of get this finalized by end of day.
254 00:23:39.450 ⇒ 00:23:44.119 Awaish Kumar: Okay, so what is, like, you already filled this data access thing.
255 00:23:44.120 ⇒ 00:23:48.379 Samuel Roberts: I just put that… yeah, but I can… I’ll… we can consider that not… there’s just a draft for now.
256 00:23:49.180 ⇒ 00:23:51.109 Awaish Kumar: Okay, you have put that in, okay.
257 00:23:51.110 ⇒ 00:23:57.640 Samuel Roberts: Yeah, I just added that based on what we had sort of talked through about getting data out, but if we don’t need to do that, it’s not important, right?
258 00:23:58.270 ⇒ 00:23:59.600 Awaish Kumar: So, yeah, okay.
259 00:23:59.760 ⇒ 00:24:03.450 Awaish Kumar: Let me know, Sam, once you have some answers.
260 00:24:03.790 ⇒ 00:24:07.429 Awaish Kumar: And then, we can also let me know about Slack.
261 00:24:07.820 ⇒ 00:24:18.059 Awaish Kumar: If there’s no direct integration, then if you need any pipeline or whatever, and basically what you’re going… I think it’s best that you should use S3, because…
262 00:24:18.830 ⇒ 00:24:24.540 Awaish Kumar: it’s native to Gemini, then if the Slack messages are in somewhere in S3,
263 00:24:25.110 ⇒ 00:24:33.089 Awaish Kumar: I think Gemini can read it from S3 as well. There must be some integration with S3. Sorry, not S3, but the Google Cloud Storage.
264 00:24:35.280 ⇒ 00:24:36.210 Samuel Roberts: Oh, okay.
265 00:24:36.780 ⇒ 00:24:44.140 Awaish Kumar: Yeah, we can put, like, Slack messages as a TXT files into Google Cloud Storage, and maybe then it can read from there, or something like that.
266 00:24:45.840 ⇒ 00:24:48.420 Samuel Roberts: Okay. I’ll spike on that and get back to you.
267 00:24:48.760 ⇒ 00:24:51.389 Samuel Roberts: Okay. For the other stuff, and then we’ll figure out that. Okay.
268 00:24:52.370 ⇒ 00:24:53.590 Awaish Kumar: Okay, thank you.
269 00:24:53.900 ⇒ 00:24:54.710 Samuel Roberts: Alright. Oof.
270 00:24:54.900 ⇒ 00:24:55.670 Pranav: Thank you, guys.
271 00:24:56.360 ⇒ 00:24:57.000 Samuel Roberts: Alrighty.