Meeting Title: Brainforge Command Center Tech Sync Date: 2026-04-14 Meeting participants: Pranav Narahari, Samuel Roberts, Awaish Kumar
WEBVTT
1 00:00:17.370 ⇒ 00:00:18.380 Samuel Roberts: Morning.
2 00:00:19.180 ⇒ 00:00:20.720 Pranav Narahari: Good morning, how you feeling?
3 00:00:21.110 ⇒ 00:00:29.160 Samuel Roberts: Better, but not… not quite 100%. I think it’s still just exhaustion, probably more than anything, but…
4 00:00:29.160 ⇒ 00:00:33.330 Pranav Narahari: Yeah, that’s fair. I just flew back last night, too.
5 00:00:33.330 ⇒ 00:00:34.380 Samuel Roberts: Oh, who were you?
6 00:00:34.380 ⇒ 00:00:34.920 Pranav Narahari: Oh.
7 00:00:35.080 ⇒ 00:00:36.040 Pranav Narahari: Huh? What’s that?
8 00:00:36.040 ⇒ 00:00:36.839 Samuel Roberts: Where were you?
9 00:00:37.230 ⇒ 00:00:39.410 Pranav Narahari: I actually went to Austin for the weekend.
10 00:00:39.410 ⇒ 00:00:40.370 Samuel Roberts: Oh, nice.
11 00:00:40.370 ⇒ 00:00:41.070 Pranav Narahari: Yeah.
12 00:00:41.890 ⇒ 00:00:46.280 Samuel Roberts: Yeah, something going… going west to east is just… it kills me every time.
13 00:00:46.610 ⇒ 00:00:49.600 Pranav Narahari: Yeah, just, like, flying at night, too, it’s just not… Oh, God.
14 00:00:49.600 ⇒ 00:01:04.159 Samuel Roberts: Yeah, yeah, and then, like, everything was delayed for me because of weather, I guess, or playing… I don’t know exactly what it was, because there was a big snowstorm out… I was out in Tahoe for a friend’s, like, bachelor-bachelorette joint party, and so I had skied, and it was, like, the first time
15 00:01:04.209 ⇒ 00:01:23.919 Samuel Roberts: all year, for the end of the season, and, like, I was just tired from that, and my body was, like, still in East Coast time, but I was trying to stay up for, you know, everyone out there, and then I got on those flights and didn’t get home till, like, 2AM, and was like, I can do this, I can do this, and then I woke up and was just, like, nauseous as hell, so… I was like, -oh. Yeah.
16 00:01:24.690 ⇒ 00:01:32.809 Samuel Roberts: Ugh, but good, good sleep yesterday was nice, so I feel a little bit better. Probably took me a few more days to get to 100%, but…
17 00:01:33.060 ⇒ 00:01:35.419 Pranav Narahari: Gotcha, gotcha. Anyways…
18 00:01:36.390 ⇒ 00:01:42.419 Pranav Narahari: Yeah, so, yeah, thanks for sending over that transcript. For whatever reason, the search wasn’t working for me, but .
19 00:01:42.420 ⇒ 00:01:46.510 Samuel Roberts: Yeah, I mean, there’s some weird… there’s some weird things there that haven’t been touched in a while, so…
20 00:01:46.510 ⇒ 00:01:49.960 Pranav Narahari: Yeah, I was able to just… She’s kind of,
21 00:01:51.120 ⇒ 00:02:04.550 Pranav Narahari: like, a review, kind of, like, the transcript as well. I just needed to do, like, some quick checks on a few things. I don’t know if you guys were able to kind of make more sense of, like, what the use case is for the command center based on that?
22 00:02:04.660 ⇒ 00:02:05.949 Pranav Narahari: Or I can kind of…
23 00:02:05.950 ⇒ 00:02:12.359 Samuel Roberts: Yeah, I mean, I went through, like, the overview a little bit. I didn’t dig into the transcript itself, because the meeting was pretty long, but…
24 00:02:13.420 ⇒ 00:02:14.050 Samuel Roberts: Okay.
25 00:02:14.050 ⇒ 00:02:18.520 Pranav Narahari: Yeah, I just use the cursor, and then kind of… Yeah, go ahead, Awish.
26 00:02:19.370 ⇒ 00:02:22.940 Awaish Kumar: Yeah, I… Went through that.
27 00:02:23.060 ⇒ 00:02:28.809 Awaish Kumar: transcript, and I… just want to understand a few things. Like, there are…
28 00:02:29.010 ⇒ 00:02:36.980 Awaish Kumar: two things that we are talking about. One is, like, adoptability, like, a quick Turn around.
29 00:02:37.370 ⇒ 00:02:40.650 Awaish Kumar: And, like, the AI adoption
30 00:02:41.270 ⇒ 00:02:45.380 Awaish Kumar: Right? In their organization, and then the command center.
31 00:02:45.760 ⇒ 00:02:47.970 Pranav Narahari: Yeah, so we’re just focused on Command Center here.
32 00:02:47.970 ⇒ 00:02:52.519 Awaish Kumar: We are talking about command center only, okay. And then for the command center.
33 00:02:52.720 ⇒ 00:02:58.639 Awaish Kumar: Do you know… do you have a list of data sources that we are prioritizing for the Phase 1?
34 00:03:00.110 ⇒ 00:03:03.319 Pranav Narahari: Yeah, so it’s, Google and Slack.
35 00:03:03.480 ⇒ 00:03:05.510 Pranav Narahari: And then from Google, we’re taking…
36 00:03:05.900 ⇒ 00:03:08.220 Awaish Kumar: Like, Google has a lot of things, right? So…
37 00:03:08.510 ⇒ 00:03:13.839 Pranav Narahari: Yeah, so from Google, we’re taking calendar data, drive activity data, and Gmail.
38 00:03:14.900 ⇒ 00:03:15.560 Awaish Kumar: Okay.
39 00:03:16.040 ⇒ 00:03:20.740 Awaish Kumar: And, Drive also includes, like, the… all the files on the drive, right?
40 00:03:21.470 ⇒ 00:03:22.260 Pranav Narahari: Yeah.
41 00:03:23.280 ⇒ 00:03:27.380 Awaish Kumar: Organization-wide, okay. And the Slack only.
42 00:03:28.230 ⇒ 00:03:28.900 Pranav Narahari: Yup.
43 00:03:29.610 ⇒ 00:03:32.759 Awaish Kumar: And it also includes emails from Gmail?
44 00:03:32.960 ⇒ 00:03:33.610 Pranav Narahari: Yeah.
45 00:03:33.880 ⇒ 00:03:34.410 Awaish Kumar: Okay.
46 00:03:34.790 ⇒ 00:03:43.050 Awaish Kumar: And then I want to understand… okay, then now it’s a question for maybe more, well, some, or…
47 00:03:43.620 ⇒ 00:03:48.809 Awaish Kumar: Is it still the… in this… like, how… how you are looking… like, is there any requirement change?
48 00:03:49.090 ⇒ 00:03:50.950 Awaish Kumar: Since we…
49 00:03:51.300 ⇒ 00:03:55.370 Pranav Narahari: I think so from yesterday’s call, things just got a little bit clearer, which I think is good.
50 00:03:55.370 ⇒ 00:03:59.739 Samuel Roberts: Can you… can you give, like, an overview real quick? Yeah, summary here, yeah.
51 00:03:59.740 ⇒ 00:04:05.710 Pranav Narahari: or, like, technical, because I think it’ll inform the technical. Basically what…
52 00:04:05.950 ⇒ 00:04:13.760 Pranav Narahari: Danny’s using it for are questions like the following, like, what are the largest three projects that the company’s working on right now?
53 00:04:14.180 ⇒ 00:04:19.379 Pranav Narahari: Things… things like that, you know?
54 00:04:20.089 ⇒ 00:04:24.769 Pranav Narahari: where is the XYZ project in terms of status?
55 00:04:24.980 ⇒ 00:04:28.689 Pranav Narahari: These are specific prompts that he was using, and…
56 00:04:28.690 ⇒ 00:04:29.270 Samuel Roberts: Okay, good.
57 00:04:29.270 ⇒ 00:04:30.390 Pranav Narahari: And so…
58 00:04:31.880 ⇒ 00:04:51.570 Pranav Narahari: those are pretty open-ended questions, right? And so they require a lot of data coming in. And so, what he was noticing with the current implementation is that sometimes it just, like, thinks for a while, and then just stops thinking. Probably runs out of tokens, is my concern. Oh, sure, okay. That’s something that we should…
59 00:04:51.870 ⇒ 00:04:55.040 Pranav Narahari: That’s something that just can’t happen, I guess,
60 00:04:56.660 ⇒ 00:05:08.940 Pranav Narahari: And there’s a few different features that we can, like, dig into this. You know, we can have, like, a shallow search, you know, if they want something quicker. We can have, like, a button that says, A, dive deeper.
61 00:05:09.820 ⇒ 00:05:11.940 Pranav Narahari: He was also mentioning how, like.
62 00:05:12.210 ⇒ 00:05:17.060 Pranav Narahari: He would like to have a little bit more of a hands-on approach.
63 00:05:17.160 ⇒ 00:05:20.749 Pranav Narahari: On which source gets dived into first.
64 00:05:21.350 ⇒ 00:05:22.370 Samuel Roberts: Mmm, okay.
65 00:05:22.660 ⇒ 00:05:37.819 Pranav Narahari: So, like, he was… I don’t know how important that is, if we’re just going to be able to dive into everything anyways, but he was, like, noticing, like, for… hey, if things are gonna be timing out, I don’t want all the time being spent on searching Google when most of the context is gonna be in Slack.
66 00:05:38.100 ⇒ 00:05:39.290 Samuel Roberts: Right.
67 00:05:39.900 ⇒ 00:05:43.569 Pranav Narahari: So, I think it’s worth, like, giving that call another listen.
68 00:05:43.570 ⇒ 00:05:44.960 Samuel Roberts: Yeah, yeah, I’ll go through it later.
69 00:05:44.960 ⇒ 00:05:56.449 Pranav Narahari: Like, but yeah, for this call, you know, we can talk a little bit about it, and then maybe when you guys, like, write the technical approach out, or just refine our current technical approach, you guys can just give that another listen.
70 00:05:56.640 ⇒ 00:06:06.079 Pranav Narahari: But… Yeah, so, does that make more sense? Like, I can answer more questions as well.
71 00:06:08.050 ⇒ 00:06:13.689 Samuel Roberts: So, okay, so we have, like, kind of example questions of the types of things he was asking. Is that… do we have more of those at all?
72 00:06:13.690 ⇒ 00:06:18.010 Pranav Narahari: Yeah, so he said that he’d give me, like, a list of those, I’m gonna ping him again today.
73 00:06:18.010 ⇒ 00:06:18.360 Samuel Roberts: Cool, cool.
74 00:06:18.360 ⇒ 00:06:18.730 Pranav Narahari: do that.
75 00:06:18.730 ⇒ 00:06:19.180 Samuel Roberts: Great.
76 00:06:19.180 ⇒ 00:06:21.910 Pranav Narahari: But I think we kinda get the gist either way, like…
77 00:06:21.910 ⇒ 00:06:24.989 Samuel Roberts: Yeah, yeah, yeah. I’m just thinking for testing purposes, I want to be able to…
78 00:06:24.990 ⇒ 00:06:34.049 Pranav Narahari: Yeah, yeah, so one question that he asked that he said he got a really good response from was, where’s the Health OS project in terms of status?
79 00:06:34.190 ⇒ 00:06:34.960 Samuel Roberts: Oh, nice, okay.
80 00:06:34.960 ⇒ 00:06:42.380 Pranav Narahari: one project that you got, he said you got a really good answer. You guys said it pulled the Slack context, channel by channel,
81 00:06:42.590 ⇒ 00:06:46.709 Pranav Narahari: Yeah, it was a really strong response, in his words.
82 00:06:48.220 ⇒ 00:06:56.390 Pranav Narahari: I think, one was, like, when he asked Eden Pharmacy Lead Up, Which was… a…
83 00:06:56.850 ⇒ 00:07:05.009 Pranav Narahari: A project that extended for over a year. However, it didn’t, you know, it was very sparse in terms of communication.
84 00:07:06.080 ⇒ 00:07:06.830 Samuel Roberts: Mmm, I won’t.
85 00:07:06.830 ⇒ 00:07:09.449 Pranav Narahari: For that year, it didn’t do a great job.
86 00:07:09.700 ⇒ 00:07:10.699 Samuel Roberts: Yeah, I can see that.
87 00:07:10.700 ⇒ 00:07:12.029 Pranav Narahari: It was just,
88 00:07:12.350 ⇒ 00:07:22.579 Pranav Narahari: it was just waiting. I think this is the one that timed out. And so, you know, I don’t know how much testing he did, I think maybe since he just did it in the same thread, maybe that thread just lost, like, you know.
89 00:07:22.840 ⇒ 00:07:29.970 Pranav Narahari: ran out of tokens. It’s not like we didn’t test… he didn’t test it very well, right? We didn’t get to do our own internal QA. Yeah, yeah.
90 00:07:29.970 ⇒ 00:07:33.540 Samuel Roberts: Well, now that we have some examples, we can definitely do that a lot better.
91 00:07:33.540 ⇒ 00:07:34.240 Pranav Narahari: Yeah.
92 00:07:34.390 ⇒ 00:07:35.340 Pranav Narahari: Yeah.
93 00:07:36.190 ⇒ 00:07:37.010 Pranav Narahari: Exactly.
94 00:07:37.010 ⇒ 00:07:46.300 Samuel Roberts: So, are you thinking that we need, because of, like, those long, or, like, a project like that, where it’s long and sparse, are you thinking we need to start doing
95 00:07:46.420 ⇒ 00:07:52.539 Samuel Roberts: a different, like… like, attack on the data kind of thing, where we might need to ingest some, or…
96 00:07:52.540 ⇒ 00:07:58.980 Pranav Narahari: No, I… I honestly, like… I don’t…
97 00:07:59.140 ⇒ 00:08:07.849 Pranav Narahari: fully know if we need to do that. I don’t… right? Like, I think I’d like for y’all to tell me if we need to do that.
98 00:08:07.850 ⇒ 00:08:14.239 Samuel Roberts: Yeah, because my thought is that as long as we… if we, like, tune the prompt, or maybe give it some extra tools that can do…
99 00:08:14.470 ⇒ 00:08:20.690 Samuel Roberts: kind of… search in a different way, because, like, right now, it just has the basic…
100 00:08:20.950 ⇒ 00:08:25.639 Samuel Roberts: you know, GWCLI, which I think is effectively just calling the API,
101 00:08:26.450 ⇒ 00:08:29.440 Samuel Roberts: But if we need to give it something that’s a little smarter about.
102 00:08:29.440 ⇒ 00:08:43.420 Pranav Narahari: The one missing link here that we didn’t have… that we didn’t know prior, which I think was, like, the cause of, like, confusion yesterday, was just that GWS doesn’t do a good job with service accounts.
103 00:08:43.760 ⇒ 00:08:47.999 Pranav Narahari: Or at least I haven’t been able to figure that out, and so maybe…
104 00:08:48.000 ⇒ 00:08:51.310 Samuel Roberts: Yeah, we knew that. That was that… that was those GitHub issues that we saw.
105 00:08:51.310 ⇒ 00:08:55.100 Pranav Narahari: PRs, right? I was, like, have issues for aliasing, but then…
106 00:08:55.440 ⇒ 00:09:11.799 Pranav Narahari: Utam, like, honestly, this is the first time I’m using domain-wide delegated, you know, service accounts. I don’t fully understand them. I thought you needed an alias. Utam said that for the platform, he hasn’t been using an alias for them, and he’s able to just kind of get all the information, like.
107 00:09:12.620 ⇒ 00:09:16.989 Pranav Narahari: Not on a profile level, just on the organization level.
108 00:09:17.630 ⇒ 00:09:24.509 Pranav Narahari: So, maybe it’s worth, Sam, like, you talking to him to see how he exactly did that, if, you know, you can’t figure that out yourself.
109 00:09:24.920 ⇒ 00:09:38.229 Samuel Roberts: Yeah, I mean, I think the question is if we’d want to lean into the CLI and just fix that thing that they nerfed, or if we want to just switch over to the API. And the thing about it is, if we’re only accessing a few specific…
110 00:09:38.570 ⇒ 00:09:47.299 Samuel Roberts: parts of Google, and we know that that’s gonna be, like, similar every time, then building a couple tools is not… you know, the thing about the CLI is that it uses the
111 00:09:47.480 ⇒ 00:10:05.449 Samuel Roberts: I think it’s called the Discovery API, or the Discovery Schema, or however Google does it, so that it doesn’t… it just, like, figures out what it needs to do based on the API. But if we have, like, if it’s only a few things, if we ever, like, oh, we always want to be able to access Calendar, we always want to be able to access Gmail, we always want to be able to access, like.
112 00:10:05.600 ⇒ 00:10:09.099 Samuel Roberts: docs, and that’s kind of it, then we can maybe, you know.
113 00:10:10.020 ⇒ 00:10:12.330 Samuel Roberts: Build out a couple more custom tools.
114 00:10:12.490 ⇒ 00:10:13.850 Samuel Roberts: That could do that.
115 00:10:15.420 ⇒ 00:10:18.489 Pranav Narahari: Yeah, yeah, we definitely could.
116 00:10:18.610 ⇒ 00:10:28.989 Pranav Narahari: I think, yeah, I think you can kind of assess, like, what is the technical complexity we need to go into in order to achieve this, and then, you know, it’s all gonna be based off of, like, time, right? Like…
117 00:10:29.320 ⇒ 00:10:38.569 Pranav Narahari: If either approach sounds like it works, you know, even data warehousing, of course, would work, but it’s just what makes the most sense for this project.
118 00:10:39.120 ⇒ 00:10:41.020 Samuel Roberts: Yeah, okay.
119 00:10:41.870 ⇒ 00:10:48.180 Awaish Kumar: I don’t know, like, with the CLI, when you have to read so many Word documents, Excel sheets, PDFs.
120 00:10:48.550 ⇒ 00:10:51.949 Awaish Kumar: Do you think it makes sense to just read the live data?
121 00:10:55.990 ⇒ 00:11:00.730 Samuel Roberts: live data? That’s what I think that that’s what we’re looking at. Are you thinking something else, you mean, Awash? What are you…
122 00:11:00.730 ⇒ 00:11:06.800 Awaish Kumar: My question is, when you are talking about that somebody prompts, and then it goes and
123 00:11:06.970 ⇒ 00:11:09.839 Awaish Kumar: Tries to read all the documents.
124 00:11:10.550 ⇒ 00:11:18.830 Awaish Kumar: Slack, Gmail, and everything. I’m saying that if the volume is big, if it has a lot of PDF, Google Sheets.
125 00:11:19.300 ⇒ 00:11:19.770 Samuel Roberts: Mmm.
126 00:11:20.570 ⇒ 00:11:24.379 Awaish Kumar: And the docs, and with each spreadsheet, with the…
127 00:11:24.570 ⇒ 00:11:28.229 Awaish Kumar: Multiple tabs, so… do you think it makes…
128 00:11:28.440 ⇒ 00:11:33.559 Awaish Kumar: sense, like, for AI agent, can it, like, grab everything and give an answer?
129 00:11:33.820 ⇒ 00:11:34.470 Awaish Kumar: Quickly?
130 00:11:34.470 ⇒ 00:11:46.579 Samuel Roberts: Yeah, I mean, I think that’s something we don’t have a ton of… like, some of these questions will be much lighter than others, it sounds like, based on how he’s asking them, so there’s a few options there where I think we could either,
131 00:11:46.730 ⇒ 00:11:54.589 Samuel Roberts: you know, if it’s a simple thing, then it just goes out and finds it. If it’s a more complex thing, maybe it spawns a sub-agent to, like, analyze… calendar, analyze…
132 00:11:54.690 ⇒ 00:11:58.249 Samuel Roberts: drive, and then report back. If it’s…
133 00:11:58.630 ⇒ 00:12:06.370 Samuel Roberts: really a ton of data, then maybe it isn’t the best strategy. I just… there’s a variety of questions, I think, that will be asked, so I’m not sure if it’s one size fits all, even.
134 00:12:06.370 ⇒ 00:12:11.959 Awaish Kumar: So the first question we should be answering is what the data volume looks like.
135 00:12:12.110 ⇒ 00:12:21.519 Samuel Roberts: Yeah, I think that’s where those example questions will be helpful, because we can test that against what we have now and see what’s actually going on, and if it’s a ton of data for certain questions, then maybe we need to…
136 00:12:22.040 ⇒ 00:12:32.499 Samuel Roberts: I think that overall architecture, but if most of them are, you know, quick ones for the current status kind of thing, and there’s only a few that are other ones, then maybe the sub-agent strategy might be a better one.
137 00:12:32.650 ⇒ 00:12:43.950 Samuel Roberts: That might even help for other things, too, where it might be… I hadn’t really dug too deep into that, thinking about the volume of the context window and stuff, because I figured… I wasn’t sure, you know, exactly the kinds of questions being asked, but I think if it’s something like…
138 00:12:44.060 ⇒ 00:12:55.859 Samuel Roberts: you know, a calendar sub-agent can go out and analyze the last year of whatever for a certain project and report back. That would simplify the context window. Or, not simplify, but streamline how much gets put into the main agent.
139 00:12:56.440 ⇒ 00:13:03.599 Awaish Kumar: Okay, so what is the state of the project right now? Is it already reading from Google Workspace, or not?
140 00:13:03.850 ⇒ 00:13:10.569 Samuel Roberts: Yes, it has the CLI access for a given user, correct, Pranav?
141 00:13:11.340 ⇒ 00:13:11.980 Pranav Narahari: Yep.
142 00:13:11.980 ⇒ 00:13:16.289 Samuel Roberts: So they auth themselves, and then it gets accessed, just like… just like Cursor has for us, basically.
143 00:13:16.670 ⇒ 00:13:23.840 Awaish Kumar: Okay, but then we had, I think this thing we discussed last time as well, that there would be something like domain…
144 00:13:24.090 ⇒ 00:13:33.249 Samuel Roberts: Yeah, that’s what Penal was just talking about. There’s some issues there with the CLI, and then we might want to just make a specific tool for calendar, specific tool for…
145 00:13:33.380 ⇒ 00:13:45.709 Samuel Roberts: drive and Gmail that uses the service account, or we have to look… because basically the CLI had that functionality, and they removed it at some point. And so, the question is, is it worth it to
146 00:13:46.400 ⇒ 00:13:52.699 Samuel Roberts: lean into that and just re-add it and fork it, or just make a couple custom tools? And I think if we’re only focused on
147 00:13:53.100 ⇒ 00:13:54.750 Samuel Roberts: Calendar, drive, and Gmail.
148 00:13:56.040 ⇒ 00:13:58.670 Samuel Roberts: Those custom tools will probably be pretty easy to whip together.
149 00:13:59.220 ⇒ 00:14:03.349 Awaish Kumar: Okay, so you are saying these tools means these are different agents, right? Sub-agents.
150 00:14:03.690 ⇒ 00:14:16.030 Samuel Roberts: They either are sub-agents, or they’re tools that specifically call the API from the main agent. But the sub-agents could make use of them as well, depending on how we architect it. So, like, instead of it being a… right now, the way it is, is that it’s just a…
151 00:14:16.290 ⇒ 00:14:17.159 Samuel Roberts: Excuse me.
152 00:14:17.420 ⇒ 00:14:20.860 Samuel Roberts: a tool that calls the CLI, and the CLI then…
153 00:14:21.070 ⇒ 00:14:35.229 Samuel Roberts: does the API lookups. So it’s one entry point, but if we make it three entry points, then we can probably make use of the service account faster, and then either give those to sub-agents or give them to the main agent, depending on how big a question gets asked.
154 00:14:36.920 ⇒ 00:14:37.610 Awaish Kumar: Okay.
155 00:14:37.880 ⇒ 00:14:40.139 Samuel Roberts: So, I think the other… Oh, go ahead, yeah.
156 00:14:40.650 ⇒ 00:14:44.120 Pranav Narahari: No, you guys can finish that up, because there’s another topic I wanted to talk about.
157 00:14:44.120 ⇒ 00:14:50.789 Samuel Roberts: Okay. Yeah, I think the other thought, I mean, I know Utom had mentioned dropping stuff into DuckDB, are we thinking we need to do that?
158 00:14:53.270 ⇒ 00:14:56.780 Pranav Narahari: again, like, I don’t know if that’s necessary, right? Like… Yeah.
159 00:14:57.320 ⇒ 00:15:05.839 Pranav Narahari: I can see both solutions working, right? Like, when we’re talking about the CLI, like, I think what we need to do here is just, we need to create a proof of concept.
160 00:15:06.180 ⇒ 00:15:13.599 Samuel Roberts: I agree, I agree. I think the proof of concept is faster without ingesting data, and then if they really do need more historical context, we.
161 00:15:13.600 ⇒ 00:15:16.709 Pranav Narahari: No, no, well, what I mean is that we need to…
162 00:15:17.060 ⇒ 00:15:29.799 Pranav Narahari: we need to do a test showing, like, hey, with this amount of data, we can… we’re able to use the CLI. So, like, based on the prompt that they gave, okay, let’s test if the CLI works.
163 00:15:29.800 ⇒ 00:15:35.950 Samuel Roberts: Yeah, yeah, no, I agree, I agree. That’s what I’m saying, I think if we can get these questions, see if, like, either the…
164 00:15:36.910 ⇒ 00:15:43.920 Samuel Roberts: You know, questions are small enough that it works as is, or they’re big enough and we need to do some sub-agents, or they’re too big and we need to ingest some data.
165 00:15:44.360 ⇒ 00:15:50.630 Pranav Narahari: Okay, yeah, I can get you more questions, but there’s a few questions, in there, right now.
166 00:15:50.730 ⇒ 00:15:57.579 Pranav Narahari: which were already causing some problems, I don’t know what the problems were, if it was just kind of…
167 00:15:57.900 ⇒ 00:16:06.469 Pranav Narahari: I don’t know if that means that we need to use a… we need to get a DuckDB, that means we need to stand up a, like, a data warehouse?
168 00:16:06.710 ⇒ 00:16:12.879 Pranav Narahari: Or if that means we need to just kind of… there’s something we can just do to, like, add another tool, or…
169 00:16:12.980 ⇒ 00:16:15.769 Pranav Narahari: One other thing that I was gonna say, too, is maybe just…
170 00:16:16.820 ⇒ 00:16:26.400 Pranav Narahari: the AI right now is a little bit more wide-sprawing, because it doesn’t have, like, a knowledge base to say, like, hey, these are, like, the initiatives for this quarter.
171 00:16:26.400 ⇒ 00:16:28.399 Samuel Roberts: That’s… yeah, I was thinking something similar, where either.
172 00:16:28.400 ⇒ 00:16:28.850 Pranav Narahari: Oh.
173 00:16:28.850 ⇒ 00:16:31.060 Samuel Roberts: Like, a scratch pad where it can store stuff?
174 00:16:31.260 ⇒ 00:16:31.720 Pranav Narahari: Right.
175 00:16:31.720 ⇒ 00:16:35.690 Samuel Roberts: And access, or we need to give it a little more info about the company, or both.
176 00:16:36.060 ⇒ 00:16:42.710 Pranav Narahari: And so that was another part of the conversation yesterday. We’re gonna get a… org chart?
177 00:16:42.850 ⇒ 00:16:43.660 Pranav Narahari: From… Oh, sick.
178 00:16:43.660 ⇒ 00:16:44.320 Samuel Roberts: Okay.
179 00:16:44.800 ⇒ 00:16:48.230 Pranav Narahari: But then also, I want to get from you, Sam, like.
180 00:16:48.400 ⇒ 00:16:56.879 Pranav Narahari: what is a good schema for this org chart? Like, what is… what is actually important that’s actually going to help with the direction for the AI?
181 00:16:56.880 ⇒ 00:17:03.940 Samuel Roberts: The org chart and the, like, projects are probably what’s most helpful to give it initial context, so when he just asks about
182 00:17:04.180 ⇒ 00:17:09.379 Samuel Roberts: whatever one of those projects was, I forget what the first one you mentioned was, but,
183 00:17:09.569 ⇒ 00:17:13.880 Samuel Roberts: It probably has to do a bunch of work to figure out what he’s even talking about at the very beginning.
184 00:17:15.300 ⇒ 00:17:21.330 Samuel Roberts: And so, if we have, like, projects, and even if we can point it to anything more specific.
185 00:17:22.119 ⇒ 00:17:28.379 Samuel Roberts: So it knows, okay, this means… this project is a thing I know about, and it can then just go search, rather than having to…
186 00:17:29.070 ⇒ 00:17:31.500 Samuel Roberts: scan all the Slack channels, or the…
187 00:17:32.690 ⇒ 00:17:33.060 Awaish Kumar: I agree.
188 00:17:33.060 ⇒ 00:17:33.940 Samuel Roberts: for everything.
189 00:17:34.300 ⇒ 00:17:35.090 Samuel Roberts: I would recommend.
190 00:17:35.090 ⇒ 00:17:37.590 Awaish Kumar: Including our… our linear.
191 00:17:39.510 ⇒ 00:17:40.610 Awaish Kumar: Into this.
192 00:17:41.190 ⇒ 00:17:49.130 Awaish Kumar: Because… like, all the projects on the data side, like, the kiosk, like, Eden OS,
193 00:17:49.450 ⇒ 00:17:52.539 Awaish Kumar: He asked about Edenoise, he asked about, for example.
194 00:17:54.150 ⇒ 00:17:56.699 Awaish Kumar: smart tech projects and things, and, like.
195 00:17:57.460 ⇒ 00:18:13.300 Awaish Kumar: some of them are, like, in our scope, already in data team. Oh, okay. So, get the info from our linear, also, regarding those projects. So, there might be some that are internal to them, but at least we get, like, 25% or 30% of projects that.
196 00:18:13.300 ⇒ 00:18:15.310 Samuel Roberts: Yeah, that’s, that’s a good, that’s good.
197 00:18:15.420 ⇒ 00:18:20.690 Samuel Roberts: Yeah, I don’t have much context into that, so I think that would be… Good to go.
198 00:18:20.690 ⇒ 00:18:22.790 Awaish Kumar: It’s just a linear team, right? A client…
199 00:18:22.790 ⇒ 00:18:25.629 Samuel Roberts: Yeah, yeah, I can see what we can pull from there, maybe, and .
200 00:18:25.970 ⇒ 00:18:26.550 Awaish Kumar: Yeah.
201 00:18:27.550 ⇒ 00:18:31.730 Samuel Roberts: Cool. Okay. Was that the next thing, Pranav, or was there something else?
202 00:18:32.480 ⇒ 00:18:34.670 Pranav Narahari: Yeah, that was gonna be the other thing, okay.
203 00:18:34.780 ⇒ 00:18:39.189 Pranav Narahari: just… I want to get, like, a… and if you can kind of, like, give me a…
204 00:18:39.340 ⇒ 00:18:45.499 Pranav Narahari: Doc, which is just kind of like the schema in which you would want this, this kind of…
205 00:18:46.090 ⇒ 00:19:01.439 Pranav Narahari: type of org chart, but it’s also going to be a more descript org chart, where… where… what type of information do we want from each individual, right? Because that’s something that they are… they have, and they’re working on, actually, because they want it for other purposes.
206 00:19:01.810 ⇒ 00:19:08.750 Pranav Narahari: Danny can also give some information there on, like, he can also provide us more information on, like.
207 00:19:09.200 ⇒ 00:19:13.169 Pranav Narahari: something specific that we would need for the command center. So, like, I think…
208 00:19:13.700 ⇒ 00:19:29.689 Pranav Narahari: somebody’s maybe… in the org chart, there’s probably going to be a lot of information in there that’s not going to be useful for the command center. What would actually be useful for us is… is what I want to kind of get from you, Sam. Like, if you can kind of think about that a little bit, that’ll be… that’ll be good.
209 00:19:30.120 ⇒ 00:19:32.760 Awaish Kumar: I have a question, like, like, agents.
210 00:19:33.310 ⇒ 00:19:39.900 Awaish Kumar: are they… like, I have… I don’t have experience with that, so it might be a knife question, but it is, like, are they able to…
211 00:19:40.480 ⇒ 00:19:46.910 Awaish Kumar: get the context from a tabular data, like, structured data, better than the document, or…
212 00:19:50.310 ⇒ 00:19:57.719 Samuel Roberts: No, I think reading the document’s probably the easiest thing if it’s just text, because that has to get in there anyway. I think there’s definitely a way we could feed it tabular data.
213 00:19:57.720 ⇒ 00:20:05.109 Awaish Kumar: Well, like, the information, like, normally we have, like, a table called… called People, which can have all the…
214 00:20:05.700 ⇒ 00:20:06.290 Awaish Kumar: like…
215 00:20:06.290 ⇒ 00:20:11.670 Pranav Narahari: I think it definitely performs better with structured, right? Like, if you have, like, JSON, if you have, like.
216 00:20:11.670 ⇒ 00:20:13.220 Samuel Roberts: Yeah. Oh, my gamma.
217 00:20:13.220 ⇒ 00:20:13.940 Pranav Narahari: point the…
218 00:20:14.380 ⇒ 00:20:17.080 Pranav Narahari: It’s gonna benefit from that structure versus just…
219 00:20:17.230 ⇒ 00:20:23.960 Samuel Roberts: The issue with, like, a table is that it would have to be able to, like, know the schema and then make the SQL calls and stuff.
220 00:20:24.210 ⇒ 00:20:29.230 Awaish Kumar: Yeah, but I’m… now, Sam, I’m talking about something that we already use, like.
221 00:20:29.230 ⇒ 00:20:29.740 Samuel Roberts: Oh, okay.
222 00:20:30.400 ⇒ 00:20:43.830 Awaish Kumar: You can have IDs, names, and you can also have a manager ID, right? So, who point… who basically reports to whom, manager, and their role description. It can be, like, you know, a single table can
223 00:20:44.010 ⇒ 00:20:53.809 Awaish Kumar: Get your answers of everything regarding an employee in the company, their role description, who’s the manager, who’s the top-level manager of this line, and things like that.
224 00:20:54.100 ⇒ 00:20:58.329 Samuel Roberts: Yeah, I mean, that could be… that could be helpful. You saying we have that already, or we have the…
225 00:20:58.620 ⇒ 00:21:02.399 Awaish Kumar: We don’t have, like, we use this structure, like, normally, like, now… Got it, yeah.
226 00:21:02.400 ⇒ 00:21:02.720 Samuel Roberts: Okay.
227 00:21:02.720 ⇒ 00:21:06.369 Awaish Kumar: There was no AI, we used to use that kind of structure.
228 00:21:06.370 ⇒ 00:21:11.430 Samuel Roberts: Yeah, yeah. That could be good. I mean, I don’t know, are they… so, Pranav, you said that they’re already putting something together?
229 00:21:12.130 ⇒ 00:21:16.500 Pranav Narahari: Yeah, I told him I would give him, like, a schema for how we want that data.
230 00:21:16.630 ⇒ 00:21:22.239 Pranav Narahari: Okay. So, I was hoping to get that from you, kind of like… and not that that’s going to be, like.
231 00:21:22.810 ⇒ 00:21:26.609 Pranav Narahari: I think that’s only going to be an enhancement for accuracy, right?
232 00:21:26.610 ⇒ 00:21:27.050 Samuel Roberts: Yeah, exactly.
233 00:21:27.050 ⇒ 00:21:32.750 Pranav Narahari: And it’s also going to make the AI a little bit more directed in giving answers.
234 00:21:32.750 ⇒ 00:21:40.190 Awaish Kumar: Yeah, like, I think instead of they giving us in a Figma, you can ask them to give us… give it in an Excel sheet format.
235 00:21:40.190 ⇒ 00:21:44.820 Samuel Roberts: Yeah, something more… yeah, definitely, definitely. Something… something like that would be better than Figma, certainly.
236 00:21:45.090 ⇒ 00:21:49.239 Pranav Narahari: Yeah, so I think what we need to do is kind of tell them what fields we’re looking for then.
237 00:21:50.140 ⇒ 00:21:52.150 Pranav Narahari: So we can give them some structure.
238 00:21:52.510 ⇒ 00:21:52.900 Samuel Roberts: Okay.
239 00:21:53.260 ⇒ 00:21:55.620 Awaish Kumar: Okay, can we create an externalized,
240 00:21:56.170 ⇒ 00:22:00.059 Awaish Kumar: Excel template, and ask them to just fill in there.
241 00:22:02.780 ⇒ 00:22:12.280 Samuel Roberts: Yeah, I guess that could work. Okay, yeah, so I’ll think about exactly what might be good. Awish, do you have any other, like, resources on this kind of thing? Because I’m not as familiar with the… the data side.
242 00:22:13.160 ⇒ 00:22:18.340 Awaish Kumar: Yeah, I can fill in the fields that… that we might need. I mean, they can fill in…
243 00:22:18.340 ⇒ 00:22:24.139 Samuel Roberts: Yeah, if you could do that, I can probably just, like, supplement a little bit, maybe, with anything else, and then we can go from there.
244 00:22:24.630 ⇒ 00:22:25.220 Awaish Kumar: Okay.
245 00:22:26.650 ⇒ 00:22:31.060 Awaish Kumar: Cool. Yeah, so is there anything, like, what are the next steps for… for us here?
246 00:22:31.220 ⇒ 00:22:42.970 Pranav Narahari: Yeah, so we… I think we need two things. I don’t know if I… if an Excel is the best way to do this. I think a JSON schema makes the most sense, just because you can have, like.
247 00:22:43.080 ⇒ 00:22:45.600 Pranav Narahari: descriptions for each person as well.
248 00:22:46.300 ⇒ 00:22:51.040 Awaish Kumar: one of that, like, if we have an Excel, converting into a JSON is not a problem for us, right?
249 00:22:51.040 ⇒ 00:22:55.680 Samuel Roberts: Yeah, yeah, I mean, if it’s Excel, or it’s CSV, or it’s whatever it is, as long as it’s just, like, structured in some way, I think it’s.
250 00:22:55.680 ⇒ 00:22:58.999 Pranav Narahari: Okay, yeah, I mean, we could do that, yeah, sure. So…
251 00:22:59.710 ⇒ 00:23:00.790 Samuel Roberts: Whatever’s kind of easier for them.
252 00:23:00.790 ⇒ 00:23:01.419 Pranav Narahari: I have to put together.
253 00:23:01.420 ⇒ 00:23:05.380 Samuel Roberts: together, I think, because we can modify it to be whatever format we need.
254 00:23:06.970 ⇒ 00:23:18.260 Pranav Narahari: Okay, okay. Yeah, I think, honestly, just then a list of fields is all I need. Yeah. Right? Like, don’t even… we don’t even need to sell them and send them an Excel doc, like, I just need, like, I guess a list of fields.
255 00:23:18.420 ⇒ 00:23:20.630 Pranav Narahari: And then…
256 00:23:22.090 ⇒ 00:23:31.840 Pranav Narahari: Projects, sure, that’s… that’s an option. Like, this is also gonna be like, hey, they have it or they don’t have it, you know? It’s not… they’re not gonna just make it for us if they don’t have it.
257 00:23:32.580 ⇒ 00:23:37.470 Pranav Narahari: So, we need to kind of be a little bit more,
258 00:23:37.810 ⇒ 00:23:55.339 Pranav Narahari: creative with what documents are already existing. So, you know, quarterly plans are gonna exist. That’s gonna have a lot of the initiatives about what are the projects they’re gonna be working on, or at least high… maybe even higher level than the projects. Projects will be synthesized from those quarterly plans.
259 00:23:55.340 ⇒ 00:23:56.140 Samuel Roberts: Yeah, my thought is that.
260 00:23:56.140 ⇒ 00:23:56.780 Pranav Narahari: We’re probably here.
261 00:23:56.780 ⇒ 00:24:01.450 Samuel Roberts: Some kind of pre-processing of that, so that the agent doesn’t have to try to redo it every time.
262 00:24:03.460 ⇒ 00:24:09.989 Pranav Narahari: Yeah, yeah, yeah. So then we can do that on our end. We can just do, like, a one-time, like, extraction of that. Yeah.
263 00:24:10.980 ⇒ 00:24:27.450 Pranav Narahari: So, yeah, we can do that. So, let’s just think about… I think quarterly plans, that’s… that’s gonna be the big one, maybe the biggest one. If you guys can think of any other, you know, just think about here, like, at Brainforge, like, where does a lot of the information about the projects exist?
264 00:24:28.010 ⇒ 00:24:36.179 Pranav Narahari: it’s a little bit different, I guess, because we have, like, we build projects on a client-specific basis, so it’s just our project plan is the source of truth.
265 00:24:36.390 ⇒ 00:24:42.109 Pranav Narahari: But… Yeah, I think maybe quarterly plans are just, like, all we need.
266 00:24:43.090 ⇒ 00:24:47.050 Pranav Narahari: But if you think of anything else, I can also just, like, say, hey, if you also have
267 00:24:47.480 ⇒ 00:24:53.140 Pranav Narahari: somewhere, like, just kind of a list of all the projects that you guys are working on? Or a list of…
268 00:24:53.280 ⇒ 00:24:55.549 Pranav Narahari: Some of the projects, even that’s helpful.
269 00:24:56.040 ⇒ 00:25:06.469 Samuel Roberts: Well, that would definitely be helpful, because if we’re going to tie that to people so it knows who to, like, look at more, you know what I mean? If it can dial in a little bit more about, like, oh, he asked about this project, we definitely want to check so-and-so.
270 00:25:06.780 ⇒ 00:25:17.739 Pranav Narahari: Yeah, at the end of the day, this will all be directionally helpful, like, even if, like, we’re not able to tie, like, there’s not, like, hey, on the HealthOS project is…
271 00:25:18.220 ⇒ 00:25:22.670 Pranav Narahari: you know, Adam, or whoever, right?
272 00:25:22.860 ⇒ 00:25:39.610 Pranav Narahari: we’ll be able to see from that org chart, like, what their description of their role is. Like, if they’re… okay, if they’re part of the engineering team, we know they’re not gonna be part of marketing, initiatives, right? And so, that’s why I’m like, okay, we probably then, for the schema, we need to know…
273 00:25:39.640 ⇒ 00:25:45.159 Pranav Narahari: Along with their name, they’re also the department they’re a part of, their role.
274 00:25:45.270 ⇒ 00:25:48.289 Pranav Narahari: Things of that nature.
275 00:25:48.550 ⇒ 00:25:49.440 Pranav Narahari: Yup.
276 00:25:49.690 ⇒ 00:25:50.740 Pranav Narahari: So…
277 00:25:51.890 ⇒ 00:26:00.890 Pranav Narahari: And then, okay, so then that’s one thing we need. We just need that, like, JSON schema, Excel sheet, whatever, however you guys want to format it. I think it really just needs to be field names. Yeah.
278 00:26:01.280 ⇒ 00:26:03.729 Pranav Narahari: Secondly, we need to… and then…
279 00:26:03.890 ⇒ 00:26:10.790 Pranav Narahari: If we can do this today or tomorrow, just, like, let’s do, like, the… Proof of concept that…
280 00:26:10.900 ⇒ 00:26:22.189 Pranav Narahari: end-to-end, we can pull this data, and it’s going to be able to provide a good answer to Danny for the 3 questions, or however many questions he asked in the
281 00:26:23.090 ⇒ 00:26:26.030 Pranav Narahari: In that, in the meeting yesterday?
282 00:26:26.260 ⇒ 00:26:26.770 Samuel Roberts: Okay.
283 00:26:27.250 ⇒ 00:26:32.850 Pranav Narahari: And then I’ll also try my best to get more questions from Danny as well.
284 00:26:32.850 ⇒ 00:26:35.350 Samuel Roberts: Cool, okay, so there’s 3 I should be able to pull out of the meeting?
285 00:26:35.860 ⇒ 00:26:44.749 Pranav Narahari: I think there should be three. I see, so I use Cursor. One of the questions it says is, where’s the HealthOS project in terms of status? That’s one question he asked.
286 00:26:46.890 ⇒ 00:26:53.669 Pranav Narahari: Also, he just typed in questions and didn’t say them out loud, you know? So there’s, like, some that you’ll just see from the recording.
287 00:26:54.790 ⇒ 00:26:55.650 Samuel Roberts: I’ll talk about that.
288 00:26:57.870 ⇒ 00:27:08.709 Pranav Narahari: Yeah, and if you have, like, more questions about that, too, like, we can hop in a call, but I think I want to kind of just… because one thing we didn’t do in the beginning of this project, which kind of…
289 00:27:09.370 ⇒ 00:27:23.350 Pranav Narahari: brought us to a little bit of, like, a blocker, was just, like, how are we gonna use the service account? It’s not gonna be through the GWS CLI, which I think we thought was gonna be what we did at first. Now it’s gonna be probably creating a new tool, or it’s gonna be…
290 00:27:23.490 ⇒ 00:27:33.320 Pranav Narahari: Sam, up to you if you think it makes more sense. I think you said a couple days ago that it doesn’t, but I’ll leave it up to you, like, to fork the CLI, and then…
291 00:27:33.590 ⇒ 00:27:40.210 Samuel Roberts: Yeah, I mean, I’ll… if we’re… I think, yeah, I’ll take a look and see how much work that is compared to just having cursors spit out a couple more tools.
292 00:27:40.630 ⇒ 00:27:41.130 Pranav Narahari: Yeah.
293 00:27:41.130 ⇒ 00:27:42.000 Samuel Roberts: API.
294 00:27:42.370 ⇒ 00:27:43.220 Pranav Narahari: Yep, yep.
295 00:27:43.710 ⇒ 00:27:49.300 Samuel Roberts: Okay, then on, like, reports and theme analysis stuff,
296 00:27:50.700 ⇒ 00:27:58.580 Samuel Roberts: where… like, where are we in the process with that? Because, like, it seemed like Uten was pretty concerned about that, and I… that’s… all that stuff was, like, out in May on the milestone, so I wasn’t sure…
297 00:28:00.120 ⇒ 00:28:00.900 Samuel Roberts: Whoa.
298 00:28:01.550 ⇒ 00:28:07.460 Pranav Narahari: Yeah, I mean, that’s still next, one thing, like…
299 00:28:08.160 ⇒ 00:28:14.240 Pranav Narahari: The whole idea with, like, this chatbot, too, is that what we’re building here
300 00:28:14.740 ⇒ 00:28:21.360 Pranav Narahari: like, in terms of pulling in all the data, like, with the CLI, with, all the Slack tools, with maybe…
301 00:28:21.560 ⇒ 00:28:25.290 Pranav Narahari: the other tools that we’re going to create, that is going to…
302 00:28:26.820 ⇒ 00:28:30.930 Pranav Narahari: Bev… that’s gonna all be utilized for creating these reports.
303 00:28:30.930 ⇒ 00:28:31.550 Samuel Roberts: Right.
304 00:28:31.720 ⇒ 00:28:43.919 Pranav Narahari: Do you still feel confident about that? That was just my understanding of it, like, all of this… it’s really gonna be pretty easy to go from this to then building the reports.
305 00:28:44.260 ⇒ 00:28:47.889 Samuel Roberts: I… yeah, that was… that was my thought, and I think that’s what we had kind of outlined in the…
306 00:28:48.480 ⇒ 00:28:57.569 Samuel Roberts: And the notion was that we would… once this is, like, working in a good way where he can get information, we just would be able to run this, you know, maybe in a little bit of a…
307 00:28:57.950 ⇒ 00:29:05.990 Samuel Roberts: Whatever, loop or schedule, or whatever to, you know, spit out a report on a given… or all the projects, or whatever, you know?
308 00:29:07.050 ⇒ 00:29:12.399 Samuel Roberts: Whether or not it needs, like, some kind of scratch pad or some kind of storage for, like, context that it needs to save.
309 00:29:12.580 ⇒ 00:29:19.770 Samuel Roberts: we could do, but I don’t think… I didn’t think there would be anything that we would need to really ingest for that, either, besides what we’re doing.
310 00:29:21.050 ⇒ 00:29:25.719 Pranav Narahari: Yeah, I think with the chatbot, too, like, it gives pretty deep…
311 00:29:26.430 ⇒ 00:29:31.879 Pranav Narahari: like, even if it’s, like, a long-running chat, right, this is only gonna be utilized by Danny.
312 00:29:32.330 ⇒ 00:29:32.930 Pranav Narahari: Right?
313 00:29:32.930 ⇒ 00:29:33.390 Samuel Roberts: Right.
314 00:29:33.390 ⇒ 00:29:35.560 Pranav Narahari: So… if…
315 00:29:35.920 ⇒ 00:29:48.500 Pranav Narahari: we get, like, first do a shallow… like, we… first, like, maybe the first answer you get is going to be just very high level on, like, the product status, and maybe it’s like, okay, no, I want you to dive deeper. That can take…
316 00:29:48.700 ⇒ 00:29:56.039 Pranav Narahari: 5 minutes, you know? Like, if you do deep research on ChatGPT, you know you’re gonna get a better answer, but it takes some time.
317 00:29:56.040 ⇒ 00:29:56.550 Samuel Roberts: Definitely.
318 00:29:56.550 ⇒ 00:29:57.760 Pranav Narahari: 16 minutes.
319 00:29:58.010 ⇒ 00:29:58.510 Samuel Roberts: Yeah.
320 00:29:58.510 ⇒ 00:30:00.340 Pranav Narahari: Right? So…
321 00:30:00.900 ⇒ 00:30:07.209 Pranav Narahari: we can explain that, right? Like, and I think that’s totally fine. Like, it’s not like these are…
322 00:30:07.880 ⇒ 00:30:14.740 Pranav Narahari: I don’t think he’s gonna be asking… like, I think that’s okay, and I think that’s the only way to do it, right?
323 00:30:16.380 ⇒ 00:30:22.479 Pranav Narahari: What is the alternative… To that, if we want to get these type of insights.
324 00:30:23.960 ⇒ 00:30:24.790 Samuel Roberts: I mean, I think…
325 00:30:25.680 ⇒ 00:30:33.590 Samuel Roberts: The other end of it is just looking at, like, ingested data, and having an agent look at all that, rather than, like, pulling it fresh.
326 00:30:33.720 ⇒ 00:30:34.950 Samuel Roberts: Which I think is what…
327 00:30:35.680 ⇒ 00:30:43.900 Samuel Roberts: who was talking about with dumping into the DuckDB, or… I mean, there is not, like, a long-term angle here, so I don’t… that’s why I wasn’t sure if that was as necessary.
328 00:30:44.370 ⇒ 00:30:45.550 Awaish Kumar: Graphic.
329 00:30:46.060 ⇒ 00:30:46.670 Samuel Roberts: Go ahead.
330 00:30:46.670 ⇒ 00:30:49.639 Awaish Kumar: I think Utam is talking about more in chat.
331 00:30:50.110 ⇒ 00:30:53.609 Awaish Kumar: In-chat data that needs to be stored, like, when somebody
332 00:30:54.470 ⇒ 00:30:57.559 Awaish Kumar: You asked a question, and it already, like,
333 00:30:58.030 ⇒ 00:31:03.819 Awaish Kumar: Grab, like, 50 files and give you an answer, and then you ask a follow-up.
334 00:31:04.890 ⇒ 00:31:06.329 Awaish Kumar: Then, I think even…
335 00:31:06.330 ⇒ 00:31:06.650 Samuel Roberts: You’re saying.
336 00:31:06.650 ⇒ 00:31:15.650 Awaish Kumar: Instead of this… the agent, again, going in and reading all those 50 documents, it should just read from what we already…
337 00:31:15.650 ⇒ 00:31:19.429 Pranav Narahari: That wouldn’t happen anyways. That would already be part of the chat context, right?
338 00:31:20.280 ⇒ 00:31:22.380 Awaish Kumar: Okay, I think that’s… Yeah.
339 00:31:22.380 ⇒ 00:31:29.350 Pranav Narahari: I don’t think that’s what Utam was saying, I think he was more so saying, like… for… let’s say…
340 00:31:29.590 ⇒ 00:31:37.369 Pranav Narahari: If that was just queried by somebody else, or by… in a different thread, then maybe it’s kind of like…
341 00:31:37.870 ⇒ 00:31:42.829 Pranav Narahari: we just save all… well, I think what he was also talking about was just, like, we just have, like, an ETL, like.
342 00:31:42.960 ⇒ 00:31:45.699 Pranav Narahari: Every day, or every… like, once a week.
343 00:31:45.700 ⇒ 00:31:57.489 Awaish Kumar: Yeah, but that he mentioned about DuckDB, right? DuckDB is something… just in-memory database. It won’t work… if it is working on my computer, it won’t work for you.
344 00:31:57.670 ⇒ 00:31:59.050 Awaish Kumar: So it is, like…
345 00:31:59.730 ⇒ 00:32:00.090 Pranav Narahari: Okay.
346 00:32:00.090 ⇒ 00:32:04.175 Awaish Kumar: It is not, like, for cross-sharing, right? It is for…
347 00:32:04.820 ⇒ 00:32:08.930 Awaish Kumar: For the same person, because DuckDB just works in memory.
348 00:32:09.150 ⇒ 00:32:13.459 Pranav Narahari: Okay, so that’s just, like, the same thing as enabling memory on… on Mastra.
349 00:32:13.460 ⇒ 00:32:19.630 Samuel Roberts: We could just… yeah, I think that’s… I mean, we definitely need to do that anyway, the memory side, and there’s a few other, like, stores we could use for…
350 00:32:19.980 ⇒ 00:32:22.799 Samuel Roberts: monster anyway, so I think that’s probably fine.
351 00:32:23.230 ⇒ 00:32:31.689 Samuel Roberts: Okay. Let’s look and see. DuckDB’s probably one of them, but we could also just… local storage or something in the browser if it’s an individual user or something. IndexDB, I don’t know.
352 00:32:31.870 ⇒ 00:32:38.369 Samuel Roberts: But I think the real point is that, yeah, once we turn that on, that’ll give more memory to the whole chat.
353 00:32:39.910 ⇒ 00:32:44.930 Samuel Roberts: Which would definitely enable it to do probably longer-running things. That might even be the problem that he was running into now.
354 00:32:45.070 ⇒ 00:32:47.639 Samuel Roberts: I’m not sure.
355 00:32:50.150 ⇒ 00:32:50.850 Pranav Narahari: Okay.
356 00:32:51.960 ⇒ 00:32:53.340 Pranav Narahari: Yeah, I guess…
357 00:32:54.820 ⇒ 00:33:01.780 Pranav Narahari: then based on that, like, I think you’ll also see just how long it takes for some of these queries to run, how much information it’s pulling.
358 00:33:01.890 ⇒ 00:33:10.319 Pranav Narahari: I think you’ll get a better understanding of, like, okay, does it just make sense for us to have this information in a data warehouse?
359 00:33:12.480 ⇒ 00:33:19.239 Pranav Narahari: Yeah, I’ll leave that up to you, too. Like, I think either, like, these solutions, in theory, both work, right?
360 00:33:19.240 ⇒ 00:33:19.760 Samuel Roberts: Right, right.
361 00:33:19.760 ⇒ 00:33:31.059 Pranav Narahari: What we need to see now is, like, in practice, how are they gonna work? And so, like, instead of just, like, trying to work on it and then shipping it to the client and seeing if it works for them, like, we should do a POC internally.
362 00:33:31.160 ⇒ 00:33:33.840 Pranav Narahari: And then make a decision, like, based on, like, that.
363 00:33:33.940 ⇒ 00:33:35.240 Pranav Narahari: that information.
364 00:33:35.730 ⇒ 00:33:43.829 Samuel Roberts: Yeah, exactly, and with the test questions, we’ll be able to do that, I think. Okay. We’ll have these three, and then hopefully it’ll give us some more, and we can actually, like, run some.
365 00:33:44.040 ⇒ 00:33:45.560 Samuel Roberts: Evaluations on them.
366 00:33:46.010 ⇒ 00:33:46.630 Pranav Narahari: Okay.
367 00:33:48.930 ⇒ 00:33:49.660 Pranav Narahari: Okay, cool.
368 00:33:49.660 ⇒ 00:33:50.220 Samuel Roberts: Cool.
369 00:33:50.510 ⇒ 00:33:58.980 Pranav Narahari: So Sam, are you gonna take point on that? You’ll kind of work with these three questions, see, like, what the output is, and then…
370 00:33:58.980 ⇒ 00:34:02.980 Samuel Roberts: Yeah, can you, do I have the link to his, the one he’s authenticated with?
371 00:34:03.690 ⇒ 00:34:06.390 Pranav Narahari: Yeah, it’s in, it’s in GCP.
372 00:34:06.970 ⇒ 00:34:08.090 Pranav Narahari: The three of them.
373 00:34:08.090 ⇒ 00:34:10.509 Samuel Roberts: Okay, do you know which, like, are they named right? Well, I know it’.
374 00:34:10.510 ⇒ 00:34:13.479 Pranav Narahari: Yeah, it’s, commandcenter-Danny.
375 00:34:13.840 ⇒ 00:34:17.979 Samuel Roberts: Okay, just making sure it wasn’t some random string. Okay, cool. Yeah, I’ll pull that up and…
376 00:34:17.989 ⇒ 00:34:25.369 Awaish Kumar: So, like, where is the… Where is… we are, like, developing this? Is there any repository or something?
377 00:34:25.830 ⇒ 00:34:32.329 Pranav Narahari: Yeah, it’s, Command Center, and it’s, like, eden-command-center in GitHub.
378 00:34:33.199 ⇒ 00:34:35.649 Awaish Kumar: Okay, can you admit to that ripple?
379 00:34:36.860 ⇒ 00:34:40.820 Pranav Narahari: Yeah, I think it’s, just shared with everybody, but yeah, I can see.
380 00:34:42.050 ⇒ 00:34:43.010 Pranav Narahari: You don’t see that right?
381 00:34:43.260 ⇒ 00:34:44.980 Awaish Kumar: I don’t see it.
382 00:34:45.159 ⇒ 00:34:45.979 Pranav Narahari: Oh, okay.
383 00:34:46.340 ⇒ 00:34:49.009 Samuel Roberts: Yeah, there’s some weird permission stuff in GitHub.
384 00:34:49.010 ⇒ 00:34:50.400 Pranav Narahari: Oh, really? Okay, I didn’t know that.
385 00:34:50.400 ⇒ 00:34:54.459 Samuel Roberts: Yeah, well, there’s… it’s like, people are on teams, and different teams are…
386 00:34:54.760 ⇒ 00:34:58.070 Samuel Roberts: Let me see who has access to this…
387 00:35:03.180 ⇒ 00:35:04.180 Awaish Kumar: Troy.
388 00:35:09.930 ⇒ 00:35:10.780 Samuel Roberts: Okay.
389 00:35:10.940 ⇒ 00:35:11.659 Samuel Roberts: There we go.
390 00:35:13.300 ⇒ 00:35:14.269 Samuel Roberts: Sorry, I cut out.
391 00:35:14.920 ⇒ 00:35:16.840 Samuel Roberts: Pseudomodin to get home.
392 00:35:18.210 ⇒ 00:35:24.310 Samuel Roberts: Pranav has direct access, the AI team and the admin team. Let me just add you, Awash.
393 00:35:24.700 ⇒ 00:35:25.360 Awaish Kumar: Okay.
394 00:35:28.110 ⇒ 00:35:34.760 Awaish Kumar: Super, and… Okay, and in the one pass… Where can I…
395 00:35:35.130 ⇒ 00:35:37.769 Awaish Kumar: Which email we are using to access it?
396 00:35:40.270 ⇒ 00:35:40.750 Pranav Narahari: Or…
397 00:35:41.010 ⇒ 00:35:43.979 Samuel Roberts: This GitHub is our… oh, sorry, for… what are you asking about?
398 00:35:43.980 ⇒ 00:35:46.119 Awaish Kumar: Gemini, Gemini, I’m asking about Gemini.
399 00:35:46.340 ⇒ 00:35:50.520 Samuel Roberts: Oh, Gemini, it’s, Pranav has a… Eden…
400 00:35:50.930 ⇒ 00:35:52.520 Awaish Kumar: Okay, okay, okay.
401 00:35:52.520 ⇒ 00:35:53.719 Samuel Roberts: It should be a 1Password.
402 00:35:53.720 ⇒ 00:35:54.499 Awaish Kumar: Yeah, I can see that.
403 00:35:54.500 ⇒ 00:35:56.220 Pranav Narahari: For Gemini?
404 00:35:56.610 ⇒ 00:35:58.100 Pranav Narahari: What do you mean for Gemini?
405 00:35:58.630 ⇒ 00:36:00.130 Awaish Kumar: Is it for, like…
406 00:36:01.950 ⇒ 00:36:03.499 Samuel Roberts: So the whole Google console.
407 00:36:03.500 ⇒ 00:36:05.519 Pranav Narahari: Oh, for Google Chronicle, okay, got you.
408 00:36:06.160 ⇒ 00:36:10.140 Samuel Roberts: So that would be how… yeah, I can… Sure, well, yeah, I…
409 00:36:10.380 ⇒ 00:36:15.189 Samuel Roberts: Yeah, it’s there, it’s calling Gemini through, the app is also running on Google Cloud Run.
410 00:36:16.020 ⇒ 00:36:18.370 Samuel Roberts: So it’s all… it’s all managed there.
411 00:36:19.270 ⇒ 00:36:20.450 Awaish Kumar: Okay, thank you.
412 00:36:20.850 ⇒ 00:36:21.410 Samuel Roberts: Yep.
413 00:36:21.980 ⇒ 00:36:23.770 Samuel Roberts: Anything else, then?
414 00:36:25.680 ⇒ 00:36:28.940 Pranav Narahari: No, I think I’m pretty good.
415 00:36:28.940 ⇒ 00:36:29.580 Samuel Roberts: Okay.
416 00:36:29.580 ⇒ 00:36:35.379 Pranav Narahari: Yeah. What do you think about… how fast do you think you could be able to spin that up, Sam?
417 00:36:36.840 ⇒ 00:36:46.090 Samuel Roberts: Let me see… hold on… I have, meeting later…
418 00:36:46.390 ⇒ 00:36:51.049 Samuel Roberts: Another meeting later, and an interview, I was like…
419 00:36:51.590 ⇒ 00:36:53.250 Samuel Roberts: They said no? I don’t know.
420 00:36:56.950 ⇒ 00:36:59.360 Samuel Roberts: Okay, I’m not sure, it might be…
421 00:36:59.840 ⇒ 00:37:12.309 Samuel Roberts: well, let me see. I’m not sure… the person for this interview declined the invite, so I don’t know if that’s happening or not now, but I’ll have to figure that out. But hopefully, midday today, I can get something going where I can at least spin up the…
422 00:37:12.690 ⇒ 00:37:14.669 Samuel Roberts: Test those questions, see the stuff.
423 00:37:14.820 ⇒ 00:37:21.170 Samuel Roberts: Maybe make some changes with the CLI and the service account? I don’t know yet.
424 00:37:22.950 ⇒ 00:37:25.539 Pranav Narahari: Yeah, like, I mean, I think you could even…
425 00:37:25.540 ⇒ 00:37:27.040 Samuel Roberts: Mike, what’s… yeah, go ahead.
426 00:37:27.040 ⇒ 00:37:31.450 Pranav Narahari: Just, yeah, like, creating your own branch and just getting things tested locally, like…
427 00:37:31.450 ⇒ 00:37:32.170 Samuel Roberts: Yeah, that’s right.
428 00:37:32.170 ⇒ 00:37:42.340 Pranav Narahari: things that we know that we can solve, and it’s just gonna be similar for everything, right? You don’t need to deploy anything, you don’t need to do, like, test cases, it really just needs to be, like.
429 00:37:42.760 ⇒ 00:37:45.160 Pranav Narahari: Does this function end-to-end?
430 00:37:45.980 ⇒ 00:37:52.399 Samuel Roberts: Right, but it’s hard to know that without, like, their… their data, because accessing my data is not as helpful, you know?
431 00:37:52.960 ⇒ 00:37:56.640 Pranav Narahari: But, is that not all… already in,
432 00:37:56.760 ⇒ 00:37:59.860 Pranav Narahari: Like, the ENV bars? Like, can’t you just pull from that?
433 00:38:00.420 ⇒ 00:38:02.759 Samuel Roberts: Well, he authenticated to his, right?
434 00:38:04.720 ⇒ 00:38:06.749 Pranav Narahari: Oh, okay, I see what you’re saying. Okay.
435 00:38:06.750 ⇒ 00:38:11.220 Samuel Roberts: I wanted to get his up and just test a few questions and understand it, and then I can kind of work from there, maybe.
436 00:38:12.060 ⇒ 00:38:12.730 Pranav Narahari: Okay.
437 00:38:13.120 ⇒ 00:38:20.220 Samuel Roberts: But yeah, I mean, we can… I could definitely test, like, tools are working and everything, but to know whether or not it answers these questions, I think, is…
438 00:38:20.520 ⇒ 00:38:22.679 Pranav Narahari: Yeah, that’s a good point. That’s a good point.
439 00:38:22.680 ⇒ 00:38:25.219 Samuel Roberts: Where we need access.
440 00:38:25.890 ⇒ 00:38:26.650 Pranav Narahari: Yeah.
441 00:38:26.880 ⇒ 00:38:30.699 Samuel Roberts: But if I can get that service account working, it might not matter? .
442 00:38:31.770 ⇒ 00:38:36.789 Pranav Narahari: Yeah, I mean, well, testing locally, like, for the questions that need to be answered.
443 00:38:37.670 ⇒ 00:38:42.280 Pranav Narahari: Yeah, I guess you don’t really need to do that on his account, right? You just need to use the service account.
444 00:38:42.640 ⇒ 00:38:50.409 Samuel Roberts: Yeah, if I can get the service account working, then we should be good. But that’s the… yeah. Otherwise, I’ll test against his for now, and just understand what the current state is of those types of questions.
445 00:38:51.170 ⇒ 00:38:57.130 Pranav Narahari: Okay, yeah, I think what’s important is that we can test with the service account, though, like… Yeah, yeah.
446 00:38:57.130 ⇒ 00:38:59.789 Samuel Roberts: That’s why, yeah, I need to get that working, I think.
447 00:39:00.220 ⇒ 00:39:00.620 Pranav Narahari: Okay.
448 00:39:01.000 ⇒ 00:39:07.800 Samuel Roberts: But I want to just see… I need to see the… I want to see the current… I’ll run those three questions, and then whatever other ones he shares, just to understand what the current state is.
449 00:39:09.350 ⇒ 00:39:10.480 Pranav Narahari: Okay, yeah, yeah, sounds good.
450 00:39:10.480 ⇒ 00:39:13.640 Samuel Roberts: is, and then I’ll develop the service account locally, so. Okay.
451 00:39:13.640 ⇒ 00:39:14.250 Pranav Narahari: Perfect.
452 00:39:17.300 ⇒ 00:39:18.180 Samuel Roberts: Alrighty.
453 00:39:18.360 ⇒ 00:39:19.280 Samuel Roberts: Anything else?
454 00:39:19.280 ⇒ 00:39:21.259 Pranav Narahari: Oh, what link are you talking about?
455 00:39:22.240 ⇒ 00:39:24.180 Awaish Kumar: Yeah, GitHub repo link.
456 00:39:24.690 ⇒ 00:39:28.759 Samuel Roberts: Oh, sorry, I just added it to you, did you, get a notification or anything?
457 00:39:29.620 ⇒ 00:39:34.020 Awaish Kumar: Yeah, sometimes I don’t, like, I don’t receive that, like, if you just send me a link, I…
458 00:39:34.400 ⇒ 00:39:36.870 Samuel Roberts: Okay, let me… let me…
459 00:39:37.150 ⇒ 00:39:38.260 Pranav Narahari: I got it right here, I got you.
460 00:39:38.260 ⇒ 00:39:41.559 Samuel Roberts: Okay, cool, yeah, you should have access now, but the link would actually be good.
461 00:39:45.910 ⇒ 00:39:50.539 Awaish Kumar: And, yeah, I sent the list of fields in the channel.
462 00:39:51.300 ⇒ 00:39:53.620 Samuel Roberts: Full name, email, department function…
463 00:39:56.150 ⇒ 00:39:57.910 Awaish Kumar: I think that should be good, right?
464 00:39:57.910 ⇒ 00:40:00.139 Samuel Roberts: That’s probably very good, yeah, I think that looks good.
465 00:40:02.650 ⇒ 00:40:04.580 Samuel Roberts: A little summary data sensitivity.
466 00:40:05.970 ⇒ 00:40:06.890 Samuel Roberts: Yeah.
467 00:40:06.890 ⇒ 00:40:07.500 Pranav Narahari: That seems cool.
468 00:40:07.500 ⇒ 00:40:08.050 Samuel Roberts: Funny.
469 00:40:08.770 ⇒ 00:40:09.420 Awaish Kumar: Sorry?
470 00:40:09.810 ⇒ 00:40:11.429 Pranav Narahari: Does, that link work? Do you have access?
471 00:40:13.410 ⇒ 00:40:17.169 Awaish Kumar: Let me… Okay, now I can see that.
472 00:40:17.520 ⇒ 00:40:18.360 Pranav Narahari: Okay, perfect.
473 00:40:23.450 ⇒ 00:40:30.209 Pranav Narahari: Yeah, so one other thing that, Utum wants is just, like, the linear tickets in, as well.
474 00:40:32.560 ⇒ 00:40:38.529 Pranav Narahari: So I think that’s gonna be having to follow once we do that POC with the service account.
475 00:40:39.140 ⇒ 00:40:46.539 Pranav Narahari: Yeah. So yeah, if you can just… Yeah, because Uten wants to, like, track this a little bit closer, just like…
476 00:40:46.670 ⇒ 00:40:57.889 Pranav Narahari: where we’re at, because… yeah. I mean, I think, yeah, he wants to move really fast. I… I think we’re moving… we are behind by one week, based on…
477 00:40:58.340 ⇒ 00:41:02.569 Pranav Narahari: our roadmap. So I think that’s why he’s pretty concerned.
478 00:41:04.660 ⇒ 00:41:10.569 Awaish Kumar: Okay. But from our conversation, I didn’t still feel any requirements for data team.
479 00:41:12.860 ⇒ 00:41:15.879 Samuel Roberts: Where are we behind? What are we missing here?
480 00:41:17.510 ⇒ 00:41:19.950 Pranav Narahari: So yesterday was the 13th, right?
481 00:41:20.290 ⇒ 00:41:23.819 Samuel Roberts: Right, so even Google Workspaces can be interacted with via chat with them, too.
482 00:41:26.110 ⇒ 00:41:28.860 Pranav Narahari: It was the entire, like, organization-wide.
483 00:41:30.630 ⇒ 00:41:32.160 Pranav Narahari: data, right?
484 00:41:36.500 ⇒ 00:41:38.859 Samuel Roberts: Doesn’t really specify that name, too, but maybe that is.
485 00:41:39.540 ⇒ 00:41:41.139 Samuel Roberts: Oh, I see what you’re saying, Danny’s workspace was the.
486 00:41:41.140 ⇒ 00:41:42.030 Pranav Narahari: Again, Google.
487 00:41:42.030 ⇒ 00:41:49.189 Samuel Roberts: I see, I see what you’re saying, yeah, yeah, okay, I got you, I got you. And then, okay, yeah, I see what you’re saying. And with anonymization, isn’t it this week? Or next week?
488 00:41:49.190 ⇒ 00:41:55.159 Pranav Narahari: Yeah, so that’s why I’m not super concerned, because we did build out some of the PII redaction stuff in another.
489 00:41:55.160 ⇒ 00:41:57.030 Samuel Roberts: Yeah, that was necessary till you met.
490 00:41:57.030 ⇒ 00:42:13.909 Pranav Narahari: I also just feel like we just need to be aligned here, and then building it out should be pretty straightforward, which is for everything, you know? Just like… I think, yeah, we’re just kind of learning that we need to go a little bit further with technical approach, with also building out some code.
491 00:42:14.000 ⇒ 00:42:19.030 Pranav Narahari: Which is, like, the proof of concept that our technical approach actually is sound.
492 00:42:19.210 ⇒ 00:42:24.910 Pranav Narahari: And so, yeah, we’ll just do that now, and I think, we should get back on track.
493 00:42:26.060 ⇒ 00:42:26.600 Samuel Roberts: Okay.
494 00:42:29.180 ⇒ 00:42:29.930 Samuel Roberts: Alright, yeah, I’ll get…
495 00:42:29.930 ⇒ 00:42:30.280 Pranav Narahari: I forgot.
496 00:42:30.280 ⇒ 00:42:31.829 Samuel Roberts: That service account working today, then?
497 00:42:32.360 ⇒ 00:42:33.030 Pranav Narahari: Cool.
498 00:42:33.570 ⇒ 00:42:34.690 Pranav Narahari: Cool, cool, cool.
499 00:42:35.270 ⇒ 00:42:41.729 Pranav Narahari: Okay, guys, yeah, that’s… that’s it for me. Let me know if I can provide any more context, but I think we’re all separate right now.
500 00:42:42.610 ⇒ 00:42:43.360 Samuel Roberts: Sounds good.
501 00:42:44.010 ⇒ 00:42:45.720 Pranav Narahari: Cool. Alright. Talk soon, guys.
502 00:42:45.720 ⇒ 00:42:46.429 Samuel Roberts: Talk to you later.
503 00:42:46.710 ⇒ 00:42:47.290 Awaish Kumar: Thank you.