Meeting Title: Uttam <> Brian Date: 2024-03-28 Meeting participants: Brian Mcgough, Uttam Kumaran


WEBVTT

1 00:00:05.360 00:00:08.640 Brian McGough: Getting getting pinged whenever there’s an error on your.

2 00:03:09.800 00:03:10.913 Uttam Kumaran: Hey, Jeremy.

3 00:03:13.270 00:03:13.990 Brian McGough: Yo.

4 00:03:14.690 00:03:15.095 Uttam Kumaran: A.

5 00:03:16.010 00:03:16.830 Brian McGough: But man.

6 00:03:17.170 00:03:17.963 Uttam Kumaran: What’s up?

7 00:03:19.110 00:03:20.369 Brian McGough: Not much was

8 00:03:20.480 00:03:22.369 Brian McGough: going on with you.

9 00:03:23.830 00:03:28.602 Uttam Kumaran: Chill in. It’s been a fun week bunch of stuff going on.

10 00:03:29.620 00:03:34.960 Uttam Kumaran: yeah, I’m excited for today, though it’s been a good at the crank of a ton of stuff last 2 days. But

11 00:03:35.260 00:03:40.410 Uttam Kumaran: the team is working on stuff. I’m kind of getting a little bit out of the loop so I can focus on this

12 00:03:40.530 00:03:41.540 Uttam Kumaran: Spanish.

13 00:03:41.550 00:03:46.170 Uttam Kumaran: We have a I have a meeting with the nimbly folks next Friday.

14 00:03:47.020 00:03:52.765 Uttam Kumaran: so that gives us a little bit of time to kind of think through what maybe it’s possible to share.

15 00:03:53.930 00:03:56.440 Uttam Kumaran: But yeah.

16 00:03:58.195 00:03:58.570 Brian McGough: Gap!

17 00:04:02.470 00:04:06.350 Uttam Kumaran: So where should we begin today?

18 00:04:10.240 00:04:16.464 Brian McGough: I guess maybe where we where we might make sense to start would be like

19 00:04:17.870 00:04:20.140 Brian McGough: what is like a deliverable

20 00:04:20.519 00:04:22.299 Brian McGough: and like, how do we

21 00:04:24.354 00:04:28.029 Brian McGough: want to approach that? And then just

22 00:04:28.160 00:04:29.260 Brian McGough: knock it out.

23 00:04:30.070 00:04:31.450 Uttam Kumaran: Okay, maybe let’s

24 00:04:32.010 00:04:35.240 Uttam Kumaran: Let me share with you like a figma.

25 00:04:36.320 00:04:37.700 Brian McGough: So, yeah, I think, oh.

26 00:04:38.360 00:04:44.800 Uttam Kumaran: I don’t. This I don’t know. If I sent it to you. I probably miss sending like a bunch of links. Let me send you

27 00:04:45.560 00:04:47.289 Uttam Kumaran: a thingma Link.

28 00:04:47.910 00:04:49.450 Brian McGough: Is the nimbly AI. One.

29 00:04:49.450 00:04:50.590 Uttam Kumaran: Yeah, that.

30 00:04:50.590 00:04:55.469 Brian McGough: Yeah, it’s got like, Workspace needs financial model for budget.

31 00:04:55.610 00:05:02.033 Uttam Kumaran: Yeah, so that that should just like, maybe we could just go in there and put like, kind of talk about like the technical architecture.

32 00:05:02.560 00:05:06.296 Uttam Kumaran: Okay, yeah. And then let me also send you

33 00:05:07.530 00:05:09.150 Uttam Kumaran: Do do do

34 00:05:09.200 00:05:13.569 Uttam Kumaran: let me also send you the agent link.

35 00:06:10.490 00:06:13.289 Brian McGough: And I went to this AI meetup last night.

36 00:06:13.290 00:06:14.449 Uttam Kumaran: Oh! How was it?

37 00:06:16.132 00:06:18.059 Brian McGough: It was like.

38 00:06:18.890 00:06:27.589 Brian McGough: I have this, maybe like next to like a nightclub, like the most crowded space I’ve been in in New York City.

39 00:06:27.590 00:06:28.360 Uttam Kumaran: Thanks.

40 00:06:29.030 00:06:32.990 Brian McGough: Crazy Bro, like everybody, was packed in. It was so hot

41 00:06:33.465 00:06:34.335 Brian McGough: and like

42 00:06:34.770 00:06:36.690 Uttam Kumaran: Was it just thoughts, or what was it.

43 00:06:37.140 00:06:42.466 Brian McGough: It was literally just. I think it was I. I don’t know if they expected it to be as big as it was, or what?

44 00:06:43.650 00:06:50.280 Brian McGough: But they just looking at this link here. Agent. Hub!

45 00:06:52.090 00:06:58.018 Brian McGough: yeah, it was wild. It was just like a ton of people packed into like this tiny room and

46 00:06:58.550 00:07:00.689 Brian McGough: The talks were

47 00:07:02.110 00:07:05.943 Brian McGough: not like overwhelmingly interesting, but

48 00:07:08.160 00:07:14.089 Brian McGough: But it was cool to see people who have like ideas for things

49 00:07:15.550 00:07:17.050 Brian McGough: and like

50 00:07:17.230 00:07:20.579 Brian McGough: taking those in whatever direction they can and

51 00:07:22.650 00:07:29.310 Brian McGough: and like, there’s a lot there was like very like a lot of very different ideas. There’s like one that was like

52 00:07:29.390 00:07:32.100 Brian McGough: an AI personal shopper.

53 00:07:32.370 00:07:38.962 Brian McGough: And then there was another which was like a person who is trying to like add additional dimensions into

54 00:07:40.200 00:07:41.250 Brian McGough: like

55 00:07:43.360 00:07:46.900 Brian McGough: into like a vector databases or something

56 00:07:49.530 00:07:53.960 Brian McGough: so like and then another person who is like doing wrappers to make like

57 00:07:54.110 00:07:56.690 Brian McGough: like certain functions

58 00:07:56.830 00:08:05.442 Brian McGough: more generic like you could switch out like all you have to do like you just say like what your intention is. And then you you could like, switch out different,

59 00:08:06.670 00:08:08.420 Brian McGough: switch out different models.

60 00:08:08.500 00:08:14.249 Brian McGough: So like, because, like different models, will have different Apis for like the same thing.

61 00:08:15.060 00:08:21.331 Brian McGough: So it was, it was really interesting, just because, like, mainly because there was a lot of variety,

62 00:08:22.220 00:08:22.965 Brian McGough: and

63 00:08:24.915 00:08:27.835 Brian McGough: but yeah, it was it. It was

64 00:08:28.380 00:08:31.350 Brian McGough: definitely packed. It was pretty wild.

65 00:08:31.730 00:08:40.569 Uttam Kumaran: Nice. Yeah. A lot of the stuff I’ve been to has been like Demos. But I don’t know. Maybe like, go to a couple of my conclusion is that it’s a lot of people

66 00:08:40.710 00:08:42.809 Uttam Kumaran: chasing problems.

67 00:08:43.470 00:08:43.919 Brian McGough: Yeah.

68 00:08:45.269 00:08:48.789 Uttam Kumaran: meaning. It’s like a lot of people that are just hobbyists. But then they’re like.

69 00:08:49.009 00:08:55.319 Uttam Kumaran: Oh, yeah, I like can solve this problem. And then it’s like, either there’s no application, or it’s like

70 00:08:55.359 00:09:00.009 Uttam Kumaran: you’re really underestimating the complication. And like the demo is just like a one small slice.

71 00:09:00.309 00:09:03.369 Uttam Kumaran: So I think I think it’s I think they were good

72 00:09:03.759 00:09:06.069 Uttam Kumaran: for me to get a sense to ask people like.

73 00:09:07.039 00:09:12.409 Uttam Kumaran: Hey, like, How like, for example, like, How are you doing? Ocr, and like, get a sense of like, what technologies are good?

74 00:09:12.539 00:09:15.219 Uttam Kumaran: Yeah. But in terms of like prom generation, I think

75 00:09:15.479 00:09:17.399 Uttam Kumaran: I think a lot of the people.

76 00:09:17.419 00:09:20.859 Uttam Kumaran: They’re just like trying random stuff, you know.

77 00:09:21.100 00:09:22.779 Brian McGough: Yeah, for sure.

78 00:09:24.671 00:09:33.349 Brian McGough: Yeah, I think I you you definitely saw a lot of that. I I think it was like a little under like, under, impressed by the problems, or like, kind of

79 00:09:33.520 00:09:34.260 Brian McGough: what?

80 00:09:37.350 00:09:38.190 Brian McGough: yeah.

81 00:09:38.390 00:09:43.570 Brian McGough: there was like, there was this one guy who had he was. It was like an AI

82 00:09:43.680 00:09:47.398 Brian McGough: meeting thing. And it was basically like,

83 00:09:48.520 00:09:51.230 Brian McGough: This was probably the most advanced product

84 00:09:51.400 00:09:55.189 Brian McGough: that I saw, but he was.

85 00:09:55.370 00:09:59.279 Brian McGough: It was like a social media platform kind of thing.

86 00:09:59.500 00:10:00.710 Brian McGough: And

87 00:10:01.080 00:10:06.620 Brian McGough: it would basically like record everything that’s said in the meeting, in in, in, like a.

88 00:10:07.500 00:10:11.330 Brian McGough: you know, in like an in person meeting like that. I don’t think it necessarily has to be in person.

89 00:10:11.730 00:10:14.959 Brian McGough: and then, like creates a transcript. And like, does all this

90 00:10:15.260 00:10:17.999 Brian McGough: like prompt engineering stuff?

91 00:10:18.000 00:10:19.329 Uttam Kumaran: Oh, okay.

92 00:10:20.680 00:10:21.300 Brian McGough: Charlie.

93 00:10:21.720 00:10:28.089 Brian McGough: So yeah, you know, it was, it was definitely, but it was. It was really cool again, just like, see people

94 00:10:28.960 00:10:36.780 Brian McGough: building things and like what people were building and like, how people were talking about it, and like how people ask questions about it. And

95 00:10:38.580 00:10:45.180 Brian McGough: so so yeah, it was I? I definitely want to go to other ones. Where do you find out about about these things?

96 00:10:45.860 00:10:55.310 Uttam Kumaran: I either see people post about on Twitter or I’m in a bunch of like Austin, like. For example, there’s an Austin like

97 00:10:55.350 00:10:57.379 Uttam Kumaran: Lane chain group

98 00:10:57.460 00:11:00.009 Uttam Kumaran: that I get email updates from

99 00:11:00.900 00:11:05.249 Uttam Kumaran: like, basically, any of the vendors come to town like snowflake or data breaks.

100 00:11:05.350 00:11:09.659 Uttam Kumaran: They host some of these events. And I don’t go to the ones where it’s like

101 00:11:10.672 00:11:14.729 Uttam Kumaran: like drinks, I go to the ones where it’s like we’re doing demos.

102 00:11:16.570 00:11:18.319 Uttam Kumaran: and I find that like way

103 00:11:18.900 00:11:19.966 Uttam Kumaran: more fun.

104 00:11:21.250 00:11:22.310 Uttam Kumaran: But.

105 00:11:22.430 00:11:24.909 Brian McGough: Damn! There was a lang chain meet up

106 00:11:26.060 00:11:27.550 Brian McGough: in New York

107 00:11:30.270 00:11:31.559 Brian McGough: couple weeks ago.

108 00:11:34.310 00:11:35.190 Uttam Kumaran: I already.

109 00:11:35.580 00:11:36.300 Brian McGough: Yeah.

110 00:11:40.120 00:11:41.440 Uttam Kumaran: Yeah, those are fine.

111 00:11:41.770 00:11:49.220 Brian McGough: Yeah, I gotta I I’ll I’ll I’ll keep an eye out. I’ll you know I’m looking on Twitter more these days. But yeah, I would definitely like to go to more

112 00:11:49.810 00:11:53.719 Brian McGough: more of these Demos and and events and stuff and

113 00:11:54.270 00:11:55.780 Brian McGough: pick people’s brains.

114 00:11:56.130 00:11:56.575 Uttam Kumaran: Yeah.

115 00:11:58.020 00:12:00.220 Uttam Kumaran: So log into a.

116 00:12:00.870 00:12:01.840 Brian McGough: Yeah, I’m in.

117 00:12:02.280 00:12:07.029 Uttam Kumaran: Okay? So you should see, kind of the current setup of the nimbly prototype.

118 00:12:07.534 00:12:10.690 Uttam Kumaran: basically kind of like how I described it.

119 00:12:13.440 00:12:15.820 Brian McGough: Alright. Let’s let me take a look here.

120 00:12:22.060 00:12:22.990 Brian McGough: care

121 00:12:27.640 00:12:30.020 Brian McGough: attributes of case study.

122 00:12:31.050 00:12:32.420 Brian McGough: Da da da.

123 00:12:39.070 00:12:40.530 Brian McGough: Do you want to kind of

124 00:12:41.100 00:12:44.039 Brian McGough: walk through what these different things are.

125 00:12:44.110 00:12:45.579 Brian McGough: Maybe like, share your screen.

126 00:12:45.750 00:12:47.590 Uttam Kumaran: Yeah, let me share my screen. 1 s.

127 00:13:25.070 00:13:25.900 Uttam Kumaran: Okay.

128 00:13:28.170 00:13:28.840 Uttam Kumaran: well.

129 00:13:29.770 00:13:33.770 Uttam Kumaran: so basically, if I go back to Sigma.

130 00:13:34.861 00:13:38.580 Uttam Kumaran: kind of like what the original architecture.

131 00:13:38.980 00:13:41.620 Uttam Kumaran: and I tried to go for

132 00:13:41.980 00:13:43.670 Uttam Kumaran: was basically

133 00:13:45.240 00:13:47.749 Uttam Kumaran: trying to do this process of

134 00:13:47.980 00:13:49.970 Uttam Kumaran: collecting information.

135 00:13:51.440 00:13:52.659 Uttam Kumaran: And then

136 00:13:52.890 00:13:56.789 Uttam Kumaran: generating these tenant requirement docs of which there are

137 00:13:57.658 00:14:02.810 Uttam Kumaran: to basically like a qualitative summary of Who is the tenant

138 00:14:03.590 00:14:07.010 Uttam Kumaran: and these are similar things to

139 00:14:07.400 00:14:09.250 Uttam Kumaran: let me show you like an example.

140 00:14:10.840 00:14:11.660 Uttam Kumaran: Well.

141 00:14:17.020 00:14:18.740 Uttam Kumaran: yeah. So here’s an example.

142 00:14:19.040 00:14:23.990 Uttam Kumaran: These are like all the properties of like a tenant, basically.

143 00:14:24.510 00:14:28.580 Uttam Kumaran: So like, what are their requirements? How do they work?

144 00:14:28.800 00:14:32.669 Uttam Kumaran: Here’s like, kind of the triggers like, what triggers? A new office?

145 00:14:33.817 00:14:36.499 Uttam Kumaran: What type of office based solutions.

146 00:14:37.240 00:14:42.670 Uttam Kumaran: So all this is kind of provided to the AI. And then here’s an example.

147 00:14:43.337 00:14:46.610 Brian McGough: Via the via the user. Input right? Like, that’s the.

148 00:14:46.760 00:14:48.030 Uttam Kumaran: Yeah, exactly.

149 00:14:48.030 00:14:50.675 Brian McGough: This is what this is, the form that the.

150 00:14:50.970 00:14:53.270 Uttam Kumaran: So. So I ideally.

151 00:14:53.630 00:14:57.410 Uttam Kumaran: these get translated to questions, and an interview is had

152 00:14:57.830 00:15:01.100 Uttam Kumaran: right where, like one or many of these are

153 00:15:01.170 00:15:08.199 Uttam Kumaran: gathered in like an interview style fashion from the user. In this case, we’re kind of skipping that step

154 00:15:08.270 00:15:10.260 Uttam Kumaran: where I’m basically just putting in

155 00:15:10.330 00:15:13.149 Uttam Kumaran: information and being like produce the output.

156 00:15:13.720 00:15:15.730 Brian McGough: Oh, okay, so yeah, I see what you’re saying.

157 00:15:16.050 00:15:18.199 Uttam Kumaran: Yeah. So here’s an example of like.

158 00:15:18.240 00:15:20.769 Uttam Kumaran: we, we produced like a fake

159 00:15:20.840 00:15:25.410 Uttam Kumaran: case study about like a fake company, or I guess it’s real company. But like

160 00:15:25.490 00:15:27.179 Uttam Kumaran: this is kind of like what

161 00:15:27.800 00:15:33.259 Uttam Kumaran: a tenant rep would produce to understand like, why a company is in this

162 00:15:33.270 00:15:39.699 Uttam Kumaran: needs office space, basically, and that kind of can understand timeline, their growth

163 00:15:40.401 00:15:43.849 Uttam Kumaran: their current solutions, their size things like that.

164 00:15:44.470 00:15:45.210 Uttam Kumaran: But

165 00:15:47.570 00:15:49.360 Uttam Kumaran: there’s also

166 00:16:01.870 00:16:06.550 Uttam Kumaran: yeah. So so basically, what I did is I have a couple of text inputs.

167 00:16:06.710 00:16:10.070 Uttam Kumaran: just basically one. This is, this is a fixed input

168 00:16:10.250 00:16:16.690 Uttam Kumaran: here. I just tell it, like, here are the attributes. So in in this, in this situation, what we’re doing is actually generating 2 documents.

169 00:16:16.770 00:16:18.790 Uttam Kumaran: we’re generating one which is like

170 00:16:18.800 00:16:20.650 Uttam Kumaran: a summary of

171 00:16:20.920 00:16:27.520 Uttam Kumaran: what the tenants needs are qualitatively as well as like a quantitative summary of like what the tenants needs are.

172 00:16:27.800 00:16:37.090 Uttam Kumaran: So basically, I’m providing it. And again, really rough. I basically provided it with a ton of different attributes of like, what a case study needs to include.

173 00:16:37.370 00:16:46.919 Uttam Kumaran: I’d have text inputs to collect the company description. The employee accounts their square feet needs location and their growth, profile.

174 00:16:46.920 00:16:48.980 Brian McGough: Oh, okay, I see. I see.

175 00:16:48.980 00:16:52.580 Uttam Kumaran: And I also put the trigger just basically what triggered them.

176 00:16:52.730 00:16:55.069 Brian McGough: Wait. So what is this input?

177 00:16:55.180 00:16:56.590 Brian McGough: What is input.

178 00:16:57.080 00:17:01.239 Uttam Kumaran: Yeah. So if you go here on the top left and you click open user view.

179 00:17:01.420 00:17:02.959 Uttam Kumaran: this is actually like what

180 00:17:03.240 00:17:04.730 Uttam Kumaran: the prototype is.

181 00:17:06.130 00:17:07.849 Uttam Kumaran: So you can input all these

182 00:17:08.020 00:17:09.460 Uttam Kumaran: hit run.

183 00:17:09.520 00:17:11.810 Uttam Kumaran: all these get collected, and then.

184 00:17:11.810 00:17:14.670 Brian McGough: So so this is like creating like a ui component.

185 00:17:14.670 00:17:17.960 Uttam Kumaran: Basic. It’s creating like a form. Basically.

186 00:17:19.920 00:17:20.700 Uttam Kumaran: where does.

187 00:17:20.700 00:17:24.419 Brian McGough: From? Where do you? Where do you export the form? Where does that live?

188 00:17:24.420 00:17:26.810 Uttam Kumaran: It just lives here at this length. So you see.

189 00:17:26.819 00:17:27.389 Brian McGough: Eliza.

190 00:17:27.390 00:17:27.950 Uttam Kumaran: Left

191 00:17:28.310 00:17:28.840 Uttam Kumaran: it is, and.

192 00:17:28.840 00:17:29.650 Brian McGough: Even the

193 00:17:29.970 00:17:32.260 Brian McGough: it lives in. It lives in Eaton Hub.

194 00:17:32.260 00:17:35.059 Uttam Kumaran: Yeah. So it’s nice to do Demos like for this.

195 00:17:38.050 00:17:43.790 Uttam Kumaran: And yeah, basically, what happens is it? It concatenates everything.

196 00:17:44.466 00:17:45.580 Uttam Kumaran: I see

197 00:17:46.280 00:17:51.140 Uttam Kumaran: it clicked. It basically does. Some prompt engineering creates like this sort of dynamic

198 00:17:51.470 00:17:57.419 Uttam Kumaran: like string. And then I had 2 prompts. One is like, create the case study. Second, is

199 00:17:57.710 00:18:00.579 Uttam Kumaran: like, create a financial analysis, basically

200 00:18:01.498 00:18:06.229 Uttam Kumaran: and then I have it output those and also write it to Google, Doc.

201 00:18:06.460 00:18:09.340 Uttam Kumaran: And the Google, Doc is here.

202 00:18:12.080 00:18:13.370 Uttam Kumaran: Basically.

203 00:18:13.910 00:18:15.340 Uttam Kumaran: here’s like a

204 00:18:16.777 00:18:19.560 Uttam Kumaran: yeah. So here’s one example of like

205 00:18:20.970 00:18:24.070 Uttam Kumaran: this is for a company called Green Point partners.

206 00:18:25.750 00:18:29.920 Uttam Kumaran: it pretty much comes up with, like what their needs are.

207 00:18:31.090 00:18:32.300 Brian McGough: Have you?

208 00:18:32.300 00:18:33.000 Uttam Kumaran: Yeah.

209 00:18:33.250 00:18:36.789 Brian McGough: Sorry I was gonna ask, have you checked this for correctness?

210 00:18:37.110 00:18:37.940 Uttam Kumaran: Now.

211 00:18:38.440 00:18:43.550 Uttam Kumaran: And and honestly, I think this is probably just like someone was playing around with

212 00:18:43.570 00:18:48.730 Uttam Kumaran: putting in Green Point, putting in all the things. There’s no dials or anything. Basically

213 00:18:48.880 00:18:53.400 Uttam Kumaran: the the thing I got to again, like in just a couple of hours that I spent was

214 00:18:53.540 00:18:57.394 Uttam Kumaran: just seeing like what the outputs were without much tweaking.

215 00:18:58.100 00:19:01.569 Uttam Kumaran: The focus of this next meeting is really on this part, which is like

216 00:19:01.680 00:19:03.519 Uttam Kumaran: the financial analysis.

217 00:19:03.940 00:19:07.100 Uttam Kumaran: But basically it does a decent job of like

218 00:19:08.680 00:19:10.300 Uttam Kumaran: like, let’s look at this example.

219 00:19:12.330 00:19:14.400 Uttam Kumaran: Here’s a rent per square feet.

220 00:19:14.410 00:19:20.689 Uttam Kumaran: Annual maintenance costs it like knows how to create a financial scenario analysis.

221 00:19:20.720 00:19:22.580 Uttam Kumaran: Yeah. Does the math right?

222 00:19:22.640 00:19:25.340 Uttam Kumaran: And comes up with something reasonable.

223 00:19:25.380 00:19:27.824 Uttam Kumaran: Lex. It comes up with something that is

224 00:19:30.107 00:19:33.862 Uttam Kumaran: like it’s thorough. I don’t know whether it’s accurate

225 00:19:34.570 00:19:51.290 Uttam Kumaran: like meaning the numbers add up. There are numbers. They seem to be in range from like what I know, but there’s no like, if this isn’t drawing on many documents, right? And so an example of like a doc that we have is like, for example, this is like a market report about Austin.

226 00:19:54.304 00:19:55.549 Uttam Kumaran: And

227 00:19:55.700 00:20:02.339 Uttam Kumaran: basically what you can see here is information about like square foot prices in different areas.

228 00:20:02.510 00:20:07.709 Uttam Kumaran: like based on what industries. So this would be something great to have as context

229 00:20:08.110 00:20:11.040 Uttam Kumaran: cause, then we can say, Oh, it’s actually pulling on

230 00:20:11.110 00:20:16.149 Uttam Kumaran: real data. And then also like, if you can do a retrieval. It can say it’s coming from this, Doc.

231 00:20:17.140 00:20:19.619 Uttam Kumaran: So like the evolution.

232 00:20:20.050 00:20:21.369 Brian McGough: Yeah, no, I feel like

233 00:20:21.930 00:20:23.600 Brian McGough: they’re

234 00:20:24.440 00:20:27.329 Brian McGough: it. It could be interesting to.

235 00:20:27.956 00:20:29.210 Uttam Kumaran: Kind of.

236 00:20:29.500 00:20:33.370 Brian McGough: Parse out the different tasks that

237 00:20:33.520 00:20:43.909 Brian McGough: the agent needs to accomplish, and, like creating different like knowledge, bases with pertinent information that it can like, search through

238 00:20:44.370 00:20:46.029 Brian McGough: and help.

239 00:20:46.250 00:20:54.519 Brian McGough: and like maybe different Apis. I don’t know if we want to like, rely on something or or what is out there for, like math.

240 00:20:54.520 00:20:54.860 Uttam Kumaran: Yeah.

241 00:20:54.860 00:20:57.270 Brian McGough: You know, cause I know there’s issues with that.

242 00:20:57.270 00:21:04.910 Uttam Kumaran: Totally. Yeah. I mean again, I would have it just be able to send like a query of like, Do math to like a calculator Api, or something like that.

243 00:21:04.990 00:21:06.469 Uttam Kumaran: where it structures like

244 00:21:06.480 00:21:10.800 Uttam Kumaran: do this math. And we can basically. But again, I’ll think a lot of that

245 00:21:11.100 00:21:21.889 Uttam Kumaran: probably isn’t as necessary for the demo, but pieces that kind of needs to happen. The other thing I wanted to share was kind of like

246 00:21:22.310 00:21:27.690 Uttam Kumaran: that’s on the slides that they produced about what their plan is. One is like.

247 00:21:28.230 00:21:31.869 Uttam Kumaran: how do you do this sort of knowledge engineering? So one is like

248 00:21:32.030 00:21:34.250 Uttam Kumaran: generating synthetic data?

249 00:21:35.003 00:21:50.930 Uttam Kumaran: So the one thing that we wanted to do is like help generate synthetic data to test the loans before we go direct to clients. For example, can we take in case studies about companies and then generate basically these tenant profiles with so

250 00:21:51.310 00:21:57.410 Uttam Kumaran: like, here are the kind of different things there’s a tenant profile. There’s the tenant interview, and there’s their workplace strategy.

251 00:21:58.720 00:22:01.730 Uttam Kumaran: and then here’s kind of like what they were

252 00:22:01.770 00:22:04.080 Uttam Kumaran: thinking about training with synthetic data which is like.

253 00:22:04.110 00:22:08.160 Uttam Kumaran: here’s a bunch of tenants, generate examples, and then

254 00:22:08.560 00:22:18.329 Uttam Kumaran: kind of have examples of like what a great tenant profile is, what a great quantity of analysis is so we can do like few shot prompting or things like that.

255 00:22:20.730 00:22:25.399 Uttam Kumaran: And yeah, again, this is kind of boilerplate slides. I’m trying to think if there’s anything else

256 00:22:29.880 00:22:34.590 Uttam Kumaran: that honestly like might be basically it. For now.

257 00:22:35.105 00:22:36.630 Uttam Kumaran: So the one thing that

258 00:22:37.097 00:22:39.460 Uttam Kumaran: one thing that I wanted to see

259 00:22:40.350 00:22:42.419 Uttam Kumaran: whether it was possible

260 00:22:44.230 00:22:45.480 Uttam Kumaran: based on.

261 00:22:45.550 00:22:48.980 Uttam Kumaran: Let me just pull up that email, too.

262 00:22:59.260 00:23:01.220 Brian McGough: I’m I’m curious, like, what

263 00:23:03.510 00:23:06.790 Brian McGough: data you were kind of envisioning in

264 00:23:07.990 00:23:15.870 Brian McGough: in in like like in vector, databases to be like ragged here. Is it?

265 00:23:16.530 00:23:21.809 Brian McGough: So we need. So it’s the tenant interview. That’s like just a form.

266 00:23:23.736 00:23:26.560 Brian McGough: So we shouldn’t need any sort of

267 00:23:29.160 00:23:31.930 Brian McGough: past information, as far as I know, for that right.

268 00:23:32.820 00:23:33.380 Uttam Kumaran: Hmm.

269 00:23:34.415 00:23:38.019 Brian McGough: And then the tenant profile is that

270 00:23:39.290 00:23:40.040 Brian McGough: does that, and.

271 00:23:40.040 00:23:49.100 Uttam Kumaran: We have to produce. So you, you can imagine, like your attendant coming from office space, you go to like a Ui, and you’re basically like.

272 00:23:49.540 00:23:55.749 Uttam Kumaran: Hey, hey, I want to get office space. It goes through a tenant interview process.

273 00:23:55.750 00:23:56.340 Brian McGough: Yep.

274 00:23:57.030 00:23:58.360 Uttam Kumaran: And then finally

275 00:23:58.590 00:24:05.909 Uttam Kumaran: cause you’re asking a bunch of questions. And then finally, you’re produced with, here’s like a what a tenant profile could

276 00:24:06.440 00:24:07.060 Uttam Kumaran: look like.

277 00:24:07.060 00:24:09.789 Brian McGough: So so is the idea that.

278 00:24:10.010 00:24:17.109 Brian McGough: or or what? I I guess I’m I’m just trying to understand. Where would we? Where would like the rag come into place.

279 00:24:18.670 00:24:22.519 Uttam Kumaran: Yeah. So where the rag will come into place is here.

280 00:24:23.350 00:24:24.770 Uttam Kumaran: Live market data.

281 00:24:25.180 00:24:25.780 Brian McGough: Okay.

282 00:24:26.600 00:24:27.640 Uttam Kumaran: So

283 00:24:28.870 00:24:31.270 Uttam Kumaran: one is getting an understanding of like

284 00:24:31.670 00:24:33.529 Uttam Kumaran: market rates cost

285 00:24:33.650 00:24:42.619 Uttam Kumaran: all that we want to pull from live, accurate data. That’s probably the only place where we need to pull in external data.

286 00:24:42.780 00:24:45.400 Uttam Kumaran: But we also may want to pull in past

287 00:24:46.420 00:24:48.270 Uttam Kumaran: past stuff that we’ve

288 00:24:48.540 00:24:49.840 Uttam Kumaran: computed. Basically.

289 00:24:49.840 00:24:50.850 Brian McGough: Yeah, yeah.

290 00:24:51.120 00:24:55.140 Uttam Kumaran: In case it’s relevant. But for the most part it’s really just this, like

291 00:24:55.350 00:25:06.439 Uttam Kumaran: live data, which is like, how do you like? What are the standards for cost. Like, how do you put together this scenarios? Again? Just giving it specific specific context on how to do that?

292 00:25:09.407 00:25:13.380 Uttam Kumaran: And then basically, what we’re trying to do is like

293 00:25:13.760 00:25:20.170 Uttam Kumaran: you’re gonna get kind of 2 summaries. One is like. And this is kind of like an example of a proposal summary. Sorry the.

294 00:25:21.030 00:25:21.570 Brian McGough: It’s fun!

295 00:25:21.570 00:25:37.253 Uttam Kumaran: Nothing blurry. But basically, this is like, kind of an example of like, here, here’s a space you get. It’s on this floor. It’s this, many square feet. Here are the terms. Here’s like, here’s the proposal. And another example of that is,

296 00:25:40.780 00:25:42.200 Uttam Kumaran: yeah, here’s an example.

297 00:25:44.720 00:25:47.339 Uttam Kumaran: So basically, in this email. He’s like.

298 00:25:52.110 00:25:56.549 Uttam Kumaran: how can AI augment or replace the work of an out, an app analyst to replace

299 00:25:56.720 00:26:05.136 Uttam Kumaran: keeping the investment models Updated and responding to investor inquiries about the deal. So this is more focused on

300 00:26:05.660 00:26:15.769 Uttam Kumaran: like the landlord side of things, but still like really focused on underwriting, commercial real estate projects. And so the question for us is like.

301 00:26:17.320 00:26:22.849 Uttam Kumaran: how can can AI summarize these financial models? And can it create

302 00:26:23.070 00:26:25.839 Uttam Kumaran: like charts, graphs, and written analysis.

303 00:26:27.540 00:26:30.479 Uttam Kumaran: And here’s an example of, like one of their

304 00:26:30.650 00:26:32.900 Uttam Kumaran: one of their like deal memos.

305 00:26:33.898 00:26:36.409 Uttam Kumaran: Basically, they’re trying to.

306 00:26:36.620 00:26:40.920 Uttam Kumaran: Yeah, I think there’s I don’t know. It’s in New Braunfels. There’s like this

307 00:26:41.230 00:26:42.620 Uttam Kumaran: parcel here.

308 00:26:42.710 00:26:46.639 Uttam Kumaran: And here’s like a bunch of analysis on, like what the

309 00:26:46.730 00:26:49.419 Uttam Kumaran: return expectations are gonna be for

310 00:26:49.430 00:26:51.179 Uttam Kumaran: this parcel. Basically

311 00:26:53.240 00:27:02.489 Uttam Kumaran: so there’s like, a probably a bunch of announce in here. But again, I think that I’m less concerned about this document, more concerned with like, it’s a technology available to

312 00:27:03.020 00:27:04.679 Uttam Kumaran: candid a document

313 00:27:04.790 00:27:07.639 Uttam Kumaran: and ask questions over it. And then for us to get

314 00:27:08.250 00:27:10.309 Uttam Kumaran: like structured data out of it. You know.

315 00:27:11.760 00:27:12.400 Brian McGough: Yeah.

316 00:27:16.930 00:27:22.600 Brian McGough: Yeah. Now I’m I’m I’m understanding more and more of the used case behind the the vision.

317 00:27:23.040 00:27:26.369 Uttam Kumaran: Yeah, so I’m on. I’ll keep. I’ll keep even just like

318 00:27:27.330 00:27:29.589 Uttam Kumaran: showing you a couple of things

319 00:27:31.230 00:27:33.510 Uttam Kumaran: and let me call one other.

320 00:28:02.270 00:28:12.799 Uttam Kumaran: So this is kind of like the entire product. Basically, it’s like you have the tenant side where we’re doing this lease analysis, finding options. And then you have the landlord side, which

321 00:28:13.070 00:28:15.740 Uttam Kumaran: powers landlords to go directly to tenants.

322 00:28:15.800 00:28:22.030 Uttam Kumaran: And then there kind of was a mock up of like what the product kind of looked like, which was like this.

323 00:28:22.260 00:28:31.130 Uttam Kumaran: which is basically like you can have like a conversation it goes through and understands all of your different office needs, and then can kind of put together like.

324 00:28:31.430 00:28:39.029 Uttam Kumaran: Here’s what you require. Here’s like kind of the things that you need. Here’s an estimated cost that puts together like a financial scenario.

325 00:28:39.450 00:28:42.939 Uttam Kumaran: This was like a really really brief mock up of something. But

326 00:28:46.310 00:28:47.250 Uttam Kumaran: yeah.

327 00:28:48.730 00:28:53.079 Uttam Kumaran: The other thing that I think would be cool to share here

328 00:28:53.140 00:28:54.660 Uttam Kumaran: is.

329 00:28:55.900 00:28:57.699 Uttam Kumaran: yeah, here, where is it?

330 00:29:00.700 00:29:07.970 Uttam Kumaran: Yeah. So here’s like, kind of like the 2 outputs and kind of how they evolve. So there’s like workspace needs, which is like what you need.

331 00:29:08.020 00:29:21.580 Uttam Kumaran: basically in a workspace like space program. And then actually like a financial analysis. So this is like what you’re gonna rent like you’re like, just basically like, put together, like a financial analysis of the lease itself.

332 00:29:22.251 00:29:29.238 Uttam Kumaran: And this is all pretty standardized. So the format and the inputs are actually all standardized. The problem is, the form factor is excel.

333 00:29:30.190 00:29:37.590 Uttam Kumaran: So I don’t know whether there’s something we could do with like. That’s why I sent that link about pandas, because maybe we could create a Csv

334 00:29:38.300 00:29:43.050 Uttam Kumaran: and then export it or something. There’s also

335 00:29:43.160 00:29:46.009 Uttam Kumaran: things about like floor plan, like

336 00:29:46.370 00:29:49.789 Uttam Kumaran: again, there’s like evolutions to each of these to get deeper and deeper.

337 00:29:51.460 00:29:52.750 Uttam Kumaran: and then

338 00:29:53.050 00:29:56.920 Uttam Kumaran: couple of examples I sent to them was.

339 00:30:01.780 00:30:03.909 Uttam Kumaran: yeah. So here’s an example of like

340 00:30:03.980 00:30:09.190 Uttam Kumaran: I was. I was explaining to them, like what the converse? That question answer, flow is, gonna be for like.

341 00:30:09.350 00:30:11.710 Uttam Kumaran: Okay, how do we do a tenant interview.

342 00:30:11.720 00:30:14.480 Uttam Kumaran: and I think I even have let me see if I have.

343 00:30:17.900 00:30:18.740 Uttam Kumaran: yeah.

344 00:30:22.320 00:30:24.759 Uttam Kumaran: So this is like an example of

345 00:30:25.110 00:30:27.020 Uttam Kumaran: what a tenant interview. Looks like

346 00:30:29.410 00:30:30.944 Uttam Kumaran: it’s gonna move this to the

347 00:30:44.020 00:30:44.770 Uttam Kumaran: but

348 00:30:52.780 00:30:55.030 Uttam Kumaran: can I just add short letters.

349 00:30:57.110 00:30:57.820 Uttam Kumaran: whatever?

350 00:30:58.819 00:31:07.120 Uttam Kumaran: So this is basically just like what? So Britney, who is the CEO of this new company she worked at. Jll. Jll. Is a very famous

351 00:31:07.130 00:31:11.550 Uttam Kumaran: like commercial real estate owner and broker, and

352 00:31:11.610 00:31:14.220 Uttam Kumaran: like they do space planning and a whole ton of stuff.

353 00:31:14.390 00:31:24.419 Uttam Kumaran: So this is like an example of like she goes through and has, like many hours of conversations with the CEO, the Cfo, and asks all these questions

354 00:31:24.820 00:31:25.790 Uttam Kumaran: like.

355 00:31:26.230 00:31:32.600 Uttam Kumaran: basically goes line by line by line, answer these questions, and in her brain figures out like, What do these guys need?

356 00:31:32.630 00:31:34.469 Uttam Kumaran: That’s what we’re trying to replicate

357 00:31:35.500 00:31:36.850 Uttam Kumaran: on the tenant side.

358 00:31:39.616 00:31:42.950 Uttam Kumaran: And basically, what I was saying is that

359 00:31:45.290 00:31:51.739 Uttam Kumaran: these? That question answer format is actually can be dynamic right? Instead of like going one by one by one?

360 00:31:52.110 00:32:01.769 Uttam Kumaran: If the AI has has context of all these questions, it could ask broad things right? It could say, we have script. It generates the questions.

361 00:32:01.850 00:32:03.440 Uttam Kumaran: We get an answer

362 00:32:03.780 00:32:19.499 Uttam Kumaran: that gets put into a conversation history. And then there’s basically like a do we have like? Did that answer answer a couple of questions? Did it answer one question, what’s the next question we should ask? So even that process of asking questions does not need to be as linear.

363 00:32:19.640 00:32:21.080 Uttam Kumaran: Yeah, yes.

364 00:32:21.520 00:32:25.400 Uttam Kumaran: So this is even another area where it’s like the interview process can be

365 00:32:25.470 00:32:31.869 Uttam Kumaran: quite dynamic like, if if in one, if within one answer they they answer like 5 of these, then cool, we could skip.

366 00:32:32.281 00:32:38.388 Uttam Kumaran: The next question itself can enter involve a couple of these, or look based on who they’re talking to.

367 00:32:38.710 00:32:49.660 Brian McGough: I think there’s I think there’s way like to your point. There’s ways we can iterate and be like, Hey, come up with a way to conduct this interview, and like we’d probably want to have something.

368 00:32:51.340 00:32:56.150 Brian McGough: you know. You know, there’s there’s there’s like, always the back and forth. It’s like, how much

369 00:32:57.140 00:33:02.449 Brian McGough: agency do you want to give? Like like we could? You know. I I would probably say.

370 00:33:02.870 00:33:07.450 Brian McGough: like I would probably argue, it would be good to have something that’s

371 00:33:08.010 00:33:13.999 Brian McGough: formalized, and but also give the agent room to be like

372 00:33:15.411 00:33:18.220 Brian McGough: you know, to deviate, if

373 00:33:18.250 00:33:19.550 Brian McGough: appropriate.

374 00:33:21.610 00:33:28.700 Uttam Kumaran: Exactly. But basically, I don’t know we could. We could do. We could think about some Kpi which is like, figure out less than 5 questions or

375 00:33:29.256 00:33:34.139 Uttam Kumaran: feedback loop from somebody. But so that’s the question side of this

376 00:33:35.128 00:33:39.670 Uttam Kumaran: and then it comes to like producing these outputs, which is like.

377 00:33:40.040 00:33:43.519 Uttam Kumaran: what’s their workspace needs? And then what’s like a financial

378 00:33:43.790 00:33:47.600 Uttam Kumaran: model for like a budget? Basically, what do we want to hand over to a landlord?

379 00:33:47.940 00:33:51.750 Uttam Kumaran: And so this actually pretty much does everything that

380 00:33:52.070 00:33:53.909 Uttam Kumaran: a tenant broker would do.

381 00:33:54.230 00:33:58.009 Uttam Kumaran: and what a space planner would do. So there’s actually a couple of different

382 00:33:58.040 00:34:03.900 Uttam Kumaran: people in this process. There’s a space plan there, and there’s like a tenant rep.

383 00:34:04.130 00:34:20.342 Uttam Kumaran: There’s like, there may be the financial agent. There’s like a bunch of shit that like a normal like. For example, if I was gonna go look for office, I would not get any of this shit. I can’t afford any of this stuff I would basically like you would go into like, here’s an example of what you would go into

384 00:34:21.929 00:34:23.596 Uttam Kumaran: like if I’m searching like.

385 00:34:26.230 00:34:27.570 Uttam Kumaran: So East Austin.

386 00:34:28.659 00:34:33.979 Uttam Kumaran: I wanna I just wanna I just wanna keep hammering home. How shitty the experiences right now.

387 00:34:34.620 00:34:37.489 Uttam Kumaran: Basically, it’s like, if I look at

388 00:34:39.480 00:34:41.509 Uttam Kumaran: commercial real estate, and Austin for rent

389 00:34:43.159 00:34:44.080 Uttam Kumaran: and.

390 00:34:44.340 00:34:45.350 Brian McGough: Contact.

391 00:34:45.580 00:34:46.639 Brian McGough: somebody.

392 00:34:47.100 00:34:49.980 Uttam Kumaran: Yeah, like, let’s let’s even look

393 00:34:55.016 00:34:56.450 Uttam Kumaran: Yeah, like.

394 00:34:58.380 00:35:01.422 Uttam Kumaran: yeah. So this is right near Kuvet. Coffee on.

395 00:35:02.080 00:35:04.390 Brian McGough: Yeah, yeah, that coffee is

396 00:35:05.290 00:35:06.140 Brian McGough: strong.

397 00:35:08.580 00:35:10.619 Uttam Kumaran: Oh, yeah. Now, I want.

398 00:35:10.620 00:35:13.350 Brian McGough: Beer garden is right next to here. What’s the place?

399 00:35:14.433 00:35:15.920 Uttam Kumaran: There’s Lazarus.

400 00:35:16.080 00:35:17.170 Brian McGough: Lazarus.

401 00:35:18.280 00:35:21.010 Uttam Kumaran: Yeah dude. Austin misses you. Bro, you gotta come back.

402 00:35:21.250 00:35:25.819 Brian McGough: Dude. My one of my best friends just put a ha offer on a house in Austin.

403 00:35:25.820 00:35:27.410 Uttam Kumaran: Oh, really! Where? At?

404 00:35:27.740 00:35:33.430 Brian McGough: I don’t. Oh, it’s in East Austin. It’s like 2. 2 of my like. 4 best friends are now.

405 00:35:36.750 00:35:41.270 Uttam Kumaran: We’ll do the nice thing, Austin, if you can come, stay, can do whatever you want. So.

406 00:35:41.760 00:35:45.910 Brian McGough: Yeah. The only problem with Austin is, if I go there, I need to have a car.

407 00:35:45.910 00:35:49.319 Uttam Kumaran: Oh, for sure. Yeah, you 100% need a car.

408 00:35:50.020 00:35:50.400 Brian McGough: Yeah.

409 00:35:50.400 00:35:51.959 Uttam Kumaran: But you could, Lisa color.

410 00:35:52.470 00:35:53.070 Uttam Kumaran: Thank you.

411 00:35:53.070 00:35:57.269 Brian McGough: Oh, no, I mean, like I meant like you know, if I was gonna go for like a week or 2.

412 00:35:57.270 00:35:59.050 Uttam Kumaran: Oh, yeah, yeah, yeah. Yeah. Yeah. Yeah.

413 00:35:59.210 00:35:59.920 Uttam Kumaran: Yeah.

414 00:36:00.550 00:36:03.000 Uttam Kumaran: I would give you a second car if I had one.

415 00:36:04.068 00:36:07.300 Brian McGough: One day, one day.

416 00:36:07.300 00:36:09.505 Uttam Kumaran: Monday. I can get the second car.

417 00:36:10.193 00:36:10.940 Brian McGough: Day soon.

418 00:36:11.279 00:36:14.329 Uttam Kumaran: But this is basically what the process is dude.

419 00:36:15.400 00:36:16.989 Uttam Kumaran: And this is all you see.

420 00:36:17.060 00:36:19.119 Uttam Kumaran: So there’s no ability to qualify.

421 00:36:19.120 00:36:19.770 Brian McGough: Road.

422 00:36:19.940 00:36:21.360 Brian McGough: Didn’t you give?

423 00:36:21.700 00:36:22.520 Brian McGough: Wait!

424 00:36:23.260 00:36:25.842 Brian McGough: No, never mind. Sorry. I’m just compute. I’m confused.

425 00:36:26.370 00:36:28.540 Brian McGough: I’m confusing something. Go go ahead.

426 00:36:29.363 00:36:31.279 Uttam Kumaran: Basically like this is all

427 00:36:31.890 00:36:34.729 Uttam Kumaran: you get. So so there’s 2 things. One is

428 00:36:36.120 00:36:56.830 Uttam Kumaran: like, I think about it like a like, I don’t know. There’s a lot of processes in life where you apply to stuff, and there’s just such a lack of information exchange during the application and qualification process that everybody’s time. It’s fucking wasted. And so what happens is, let’s say, I put in a thing here. And I put in this, not only is this like? Probably not enough information?

429 00:36:57.304 00:37:10.289 Uttam Kumaran: They basically don’t collect any information about like, does my company currently have office space? What’s my growth? And my Vc. I don’t. I don’t like. I can’t really put my information in and get an understanding of like.

430 00:37:10.630 00:37:14.379 Uttam Kumaran: Do I match with the space like

431 00:37:14.740 00:37:16.629 Uttam Kumaran: I don’t know fucking big this is.

432 00:37:16.730 00:37:18.840 Uttam Kumaran: and I don’t know like

433 00:37:18.940 00:37:32.219 Uttam Kumaran: I don’t know. Do I need like? Do I need like like outlets everywhere, like I don’t know what all that shit isn’t. Again, not a lot of that information is even here. So basically, what we want to do is take like a listing like basically

434 00:37:32.390 00:37:40.030 Uttam Kumaran: combine our really sophisticated knowledge of the tenant with hopefully, like a really good understanding of the

435 00:37:40.050 00:37:43.319 Uttam Kumaran: like spaces available and then match make.

436 00:37:43.540 00:37:57.290 Uttam Kumaran: And so the process typically is like, I would submit a thing. They would go in, and then I would just go get a tour, and then they would ask you for a bunch of documents they would spend. It’d be like a it’d be complete waste of everybody’s time. Because I’m not qualified for this, I you know, and so

437 00:37:57.300 00:38:00.060 Uttam Kumaran: not only, but then also, like I would go.

438 00:38:00.130 00:38:06.210 Uttam Kumaran: I have to get a broker, and I have to go in and ask answer a bunch of questions that I have no idea about, like I don’t know.

439 00:38:06.480 00:38:12.840 Uttam Kumaran: I really don’t know that I need to understand. Like, do I need a fucking kitchen, or like.

440 00:38:12.920 00:38:15.570 Uttam Kumaran: what’s my lighting strategy like.

441 00:38:15.880 00:38:28.259 Uttam Kumaran: you know what I mean? I don’t know any of this. And so Space Planner to come in. And so basically, space planner is quite expensive. And only the people like who are renting a lot of auto space typically get that service.

442 00:38:28.938 00:38:34.009 Uttam Kumaran: Because the brokers throw it in. But that’s another added service that we want to kind of

443 00:38:34.030 00:38:38.939 Uttam Kumaran: make possible. Again, what is what is that that’s just answering this fucking questions.

444 00:38:42.260 00:38:43.133 Brian McGough: Yeah. Man.

445 00:38:44.330 00:38:45.100 Brian McGough: Yeah.

446 00:38:46.050 00:38:47.460 Uttam Kumaran: So maybe we can

447 00:38:47.900 00:38:50.540 Uttam Kumaran: like iterate a little bit on

448 00:38:50.970 00:38:54.970 Uttam Kumaran: this, which I think is like could just start to be like

449 00:38:56.160 00:38:57.390 Uttam Kumaran: the

450 00:38:57.800 00:38:59.460 Uttam Kumaran: like. What we think is like

451 00:38:59.710 00:39:05.469 Uttam Kumaran: the next steps here. But I think there’s 2 things that we want to try to do. One is

452 00:39:05.770 00:39:09.219 Uttam Kumaran: just have a data store for stuff like we talked about

453 00:39:09.530 00:39:10.500 Uttam Kumaran: right?

454 00:39:12.260 00:39:15.050 Uttam Kumaran: So let me even just like write some of this.

455 00:39:18.110 00:39:27.610 Uttam Kumaran: So one is like, just the data store. So what’s this gonna have? It’s gonna have like past interviews. It’s gonna have, like market rates.

456 00:39:31.620 00:39:33.519 Brian McGough: Feel like the market rates might.

457 00:39:34.460 00:39:35.320 Brian McGough: Yeah.

458 00:39:35.730 00:39:39.809 Brian McGough: Market rates. Maybe that’s is there like an Api, we can tap into.

459 00:39:39.810 00:40:07.789 Uttam Kumaran: Yeah, maybe there’s an Api. But again, the the basically like, we could just focus up Austin for now, cause I think we’re just using that like as an example. But like and every quarter a lot of these guides come out from the big brokers, which is just explaining, like, what’s the state of the market. So it’s pretty easy to get that and shove that in. And again, this. These don’t need to be like accurate to the dollar. And these don’t feel like real time, this data. I don’t even know whether it’s really available real time.

460 00:40:07.970 00:40:14.169 Uttam Kumaran: But a lot of people just put this out like on the end of the quarter which is like, here’s the state of the Austin real estate market, basically.

461 00:40:14.640 00:40:17.609 Brian McGough: When you say market rates, what are the actual?

462 00:40:17.630 00:40:19.070 Brian McGough: Yeah, what are?

463 00:40:19.990 00:40:25.679 Brian McGough: What is? What do you mean? What is the market rate market rate is like an insurance or or not insurance.

464 00:40:27.740 00:40:30.370 Uttam Kumaran: Yeah, let me even show you like, what.

465 00:40:30.370 00:40:33.460 Brian McGough: Average cost for real estate, or is it.

466 00:40:33.900 00:40:37.520 Uttam Kumaran: Let me show you like, what? What are those things look like?

467 00:40:44.180 00:40:47.209 Uttam Kumaran: But this is like an Austin. Q, 4 thing.

468 00:40:47.400 00:40:50.379 Uttam Kumaran: Yeah. So basically, understanding like, what’s the

469 00:40:50.450 00:40:54.639 Uttam Kumaran: average price per square feet in different areas?

470 00:40:54.840 00:40:56.399 Uttam Kumaran: What’s available

471 00:40:56.540 00:40:58.540 Uttam Kumaran: where you could see like the.

472 00:40:58.540 00:41:01.090 Brian McGough: Why is important in this use? Case.

473 00:41:01.350 00:41:03.169 Uttam Kumaran: To do the financial

474 00:41:03.940 00:41:06.589 Uttam Kumaran: financial scenario. Doc.

475 00:41:07.120 00:41:07.800 Brian McGough: Okay.

476 00:41:08.220 00:41:12.739 Uttam Kumaran: So to help them get an accurate understanding of hey, you need 5,000 square feet in East Austin.

477 00:41:12.990 00:41:13.430 Brian McGough: Yeah.

478 00:41:13.430 00:41:16.460 Uttam Kumaran: This is the market rate, right? And again, that’s we could say.

479 00:41:16.460 00:41:17.190 Brian McGough: Yeah.

480 00:41:17.190 00:41:19.669 Uttam Kumaran: You use 5,000 square feet of Class B,

481 00:41:20.110 00:41:22.790 Uttam Kumaran: it’s gonna be 5,000 times this.

482 00:41:22.790 00:41:25.390 Brian McGough: Is that consistent, like generally.

483 00:41:26.430 00:41:27.770 Uttam Kumaran: Consistent meeting.

484 00:41:28.040 00:41:32.989 Brian McGough: Like like you know, if I’m looking at apartments, I might get like

485 00:41:33.360 00:41:36.629 Brian McGough: I might find one apartment that is

486 00:41:36.850 00:41:46.989 Brian McGough: comparable to another, but is significantly more expensive or cheaper, like, you know, is there? Do, pet, do, do? Does commercial real estate

487 00:41:47.230 00:41:49.849 Brian McGough: generally follow market price?

488 00:41:51.100 00:41:58.929 Uttam Kumaran: Yes, so there’s a couple of different variables. And like, I’ll kind of explain a couple of those but one

489 00:41:59.300 00:42:06.040 Uttam Kumaran: segmented by where it is in the class, there is typically an asking rate

490 00:42:06.420 00:42:13.250 Uttam Kumaran: that is like the market rate. But there’s a lot of other levers that you could pull beyond just the price for the square feet.

491 00:42:13.250 00:42:14.190 Brian McGough: Like amenities.

492 00:42:14.487 00:42:18.650 Uttam Kumaran: A couple of examples are like, What’s the what’s the length of your lease?

493 00:42:19.069 00:42:38.539 Uttam Kumaran: How much money like like? What other do you need to modify the space a lot. So you need a lot of like treatment money. So there is a lot of other levers beyond just the what’s the price for square foot in the area. But this is like typically these docs are what the landlords go to to understand.

494 00:42:38.850 00:42:41.099 Uttam Kumaran: Like, what’s the floor? Basically.

495 00:42:41.100 00:42:42.959 Brian McGough: So if we

496 00:42:44.780 00:42:45.620 Brian McGough: okay.

497 00:42:46.020 00:42:56.549 Uttam Kumaran: So you can think about the tenant rep side is purely. Can we give the tenants an understanding of what their scenarios are going to be when they go seek office space.

498 00:42:56.690 00:42:57.740 Uttam Kumaran: Yep, yep.

499 00:42:58.703 00:42:59.236 Uttam Kumaran: cool.

500 00:42:59.770 00:43:01.579 Brian McGough: See? I, yeah. So.

501 00:43:01.580 00:43:02.330 Uttam Kumaran: Excludes.

502 00:43:02.330 00:43:04.600 Brian McGough: Here’s what you should expect.

503 00:43:04.600 00:43:05.090 Uttam Kumaran: Yes.

504 00:43:05.870 00:43:09.200 Brian McGough: And then we’re gonna give you, yeah, okay.

505 00:43:09.200 00:43:11.569 Uttam Kumaran: So we’re gonna say, look you, wanna you wanna

506 00:43:11.870 00:43:31.159 Uttam Kumaran: 10 year lease in East Austin for 5,000 square feet for room to grow by an extra 50 people with a ton of phone booths and a ton of things. Here’s basically your conservative liberal and like base case for how much you’re gonna spend that basically can look something like

507 00:43:33.190 00:43:35.460 Uttam Kumaran: the mic. Oh.

508 00:43:36.140 00:43:38.450 Uttam Kumaran: something like this, which is like.

509 00:43:39.060 00:43:42.449 Uttam Kumaran: here’s what you need. Here’s your foot. Here’s your floor like

510 00:43:42.480 00:43:44.130 Uttam Kumaran: there’s 3 scenarios.

511 00:43:44.210 00:43:47.399 Uttam Kumaran: The other things that are kind of variable are like

512 00:43:49.560 00:43:55.340 Uttam Kumaran: things on like trends. Where, again, this may be important information to layer on

513 00:43:55.620 00:43:59.460 Uttam Kumaran: as like oh! The the the agent actually knows, like

514 00:43:59.510 00:44:07.270 Uttam Kumaran: the markets. Is this a good time to buy like it becomes a. It becomes an actual representative like it can actually

515 00:44:07.310 00:44:11.699 Uttam Kumaran: have a understanding and dude. I’m telling you, then, real estate.

516 00:44:11.770 00:44:16.949 Uttam Kumaran: It’s literally just looking at these numbers and putting together like a basic ass spreadsheet.

517 00:44:17.070 00:44:19.100 Uttam Kumaran: And then a lot of it is just like.

518 00:44:19.490 00:44:33.129 Uttam Kumaran: can we? Can we get the tenant profile ready so that we can market to landlords and the the other the other really bad thing that happens is that the process of actually getting these you literally email

519 00:44:33.210 00:44:34.840 Uttam Kumaran: like a fucking?

520 00:44:36.160 00:44:36.890 Uttam Kumaran: Landlord.

521 00:44:37.070 00:44:59.119 Uttam Kumaran: And you email them with, just like, Hey, I need office space. And then there’s a bunch of things that kick off and they still haven’t qualify you. So the the goal of these guys is like, imagine you could email the landlord. So think about this. The the profile is like, Oh, you got an alert from nimbly nimble is like, we have a tenant. Here’s their profile. They’re looking for office space in your area.

522 00:44:59.190 00:45:03.130 Uttam Kumaran: Sign up and we’ll give you their their name and contact.

523 00:45:03.150 00:45:05.360 Uttam Kumaran: That’s what they’re thinking about.

524 00:45:05.800 00:45:06.300 Uttam Kumaran: Yeah.

525 00:45:06.300 00:45:06.910 Brian McGough: Mix up.

526 00:45:06.910 00:45:10.639 Uttam Kumaran: Because they never get these co pre-qualified tenants.

527 00:45:10.850 00:45:18.820 Uttam Kumaran: With this much great data, which is like all the types of spaces there need their profile and their financial profile.

528 00:45:19.030 00:45:21.820 Uttam Kumaran: That’s something that the landlord has to do.

529 00:45:22.500 00:45:24.679 Uttam Kumaran: And it takes a ton of time to do.

530 00:45:24.820 00:45:28.220 Uttam Kumaran: Yeah, yeah, no. This makes perfect connection. Basically.

531 00:45:30.110 00:45:30.790 Brian McGough: Cool.

532 00:45:32.437 00:45:44.250 Brian McGough: So I’m wondering what is the thing you know? Kind of getting back to the original point, because I appreciate that deep dive so I’m wondering what

533 00:45:46.110 00:45:47.140 Brian McGough: you know.

534 00:45:47.667 00:45:53.030 Brian McGough: I it sounds like the tenant profiler is like the big thing here, right?

535 00:45:54.090 00:45:55.619 Brian McGough: That’s like, that’s like.

536 00:45:55.650 00:46:00.819 Brian McGough: you know the main that’s like a core value. Add, and

537 00:46:01.120 00:46:02.779 Brian McGough: if we have that.

538 00:46:03.650 00:46:04.880 Brian McGough: then

539 00:46:05.906 00:46:09.393 Brian McGough: I you know I it sounds like that. Would that would

540 00:46:12.830 00:46:15.899 Brian McGough: you know, provide a lot of immediate benefit.

541 00:46:17.290 00:46:23.240 Uttam Kumaran: Yeah, I think there’s there’s like the 2 things. There’s one which is like the tenant interview process.

542 00:46:23.340 00:46:25.879 Uttam Kumaran: the creation of the qualitative

543 00:46:26.360 00:46:27.620 Uttam Kumaran: profile. Yeah.

544 00:46:27.670 00:46:29.620 Uttam Kumaran: the second thing is

545 00:46:29.630 00:46:31.649 Uttam Kumaran: the financial analysis

546 00:46:31.740 00:46:32.880 Uttam Kumaran: which is like.

547 00:46:33.030 00:46:37.790 Uttam Kumaran: can you ask people like what their revenues are? And can you put together like.

548 00:46:37.830 00:46:40.230 Uttam Kumaran: based on where they’re looking for office space.

549 00:46:40.290 00:46:43.609 Uttam Kumaran: What like example, lease terms could look like

550 00:46:44.100 00:46:50.630 Uttam Kumaran: right. But underneath both of those I think there are like primitive technologies or primitive things that we need to figure out

551 00:46:51.342 00:46:56.389 Uttam Kumaran: which can be applied to like. For example, in this, in this email, they’re asking me to think about like.

552 00:46:56.570 00:46:59.520 Uttam Kumaran: hey? Can we think about what this may look like on the landlord side?

553 00:47:00.110 00:47:00.800 Uttam Kumaran: Yep.

554 00:47:01.130 00:47:01.950 Uttam Kumaran: But

555 00:47:02.110 00:47:05.550 Uttam Kumaran: again, that I don’t think the technology is much different than what we talk about on tenant side?

556 00:47:05.880 00:47:09.910 Uttam Kumaran: Right? They’re like, can we run a fucking thing over this guy and ask some questions? Yeah.

557 00:47:09.910 00:47:11.160 Brian McGough: You mean like you mean, like.

558 00:47:11.160 00:47:11.700 Uttam Kumaran: Yeah, that.

559 00:47:12.360 00:47:19.029 Brian McGough: You you mean, like from the tenant side being like, Hey, you’re about to list something like, let me ask you a bunch of questions about it.

560 00:47:20.219 00:47:23.089 Uttam Kumaran: Yes, on the landlord side. Yeah. So, for example.

561 00:47:23.090 00:47:23.420 Brian McGough: Yeah, yeah.

562 00:47:23.420 00:47:31.240 Uttam Kumaran: So this guy, this guy gentry who’s on here is like, kind of like a Bro, he’s a big broker here in Austin. And

563 00:47:31.430 00:47:34.290 Uttam Kumaran: basically, they were like, we wanna put together a workshop to

564 00:47:34.300 00:47:48.289 Uttam Kumaran: dig into a couple of use cases in a day in a gentry’s life, which is like, here’s like one of their asset memos. Can AI augment or replace the work of an out analysis to keep investment models updated and responding to investor inquiries about the deal.

565 00:47:48.760 00:47:49.219 Brian McGough: Oh, yeah.

566 00:47:49.220 00:47:54.249 Uttam Kumaran: Sound that sounds like rag over a Pdf. With hopefully, some like

567 00:47:54.270 00:47:57.160 Uttam Kumaran: good ability to Ocr or

568 00:47:57.210 00:48:00.620 Uttam Kumaran: like, ask some financial questions and do some math on the fly.

569 00:48:00.840 00:48:03.950 Uttam Kumaran: right? So kind of the things that I sent was like.

570 00:48:04.110 00:48:11.079 Uttam Kumaran: Here’s how you can ask free text questions using pandas. Here’s some of latest and advanced. Pdf, parsing.

571 00:48:12.980 00:48:16.419 Uttam Kumaran: those are kind of the examples that I sent, and I sent those in slack.

572 00:48:16.530 00:48:17.510 Uttam Kumaran: But

573 00:48:17.700 00:48:20.129 Uttam Kumaran: yeah, I guess there’s a there’s a couple of.

574 00:48:20.350 00:48:24.426 Uttam Kumaran: I guess. Let’s try to think about how do we break this down

575 00:48:25.050 00:48:26.090 Uttam Kumaran: to

576 00:48:26.846 00:48:30.859 Uttam Kumaran: maybe we just talk purely about this angle first.

577 00:48:32.340 00:48:32.930 Brian McGough: Yeah.

578 00:48:33.641 00:48:34.979 Brian McGough: yeah. I guess. So what? I would.

579 00:48:34.980 00:48:36.670 Uttam Kumaran: So yeah, go ahead.

580 00:48:36.670 00:48:38.704 Brian McGough: What I was. What I was saying is like,

581 00:48:39.940 00:48:41.070 Brian McGough: yeah, what

582 00:48:41.180 00:48:43.150 Brian McGough: can we

583 00:48:44.350 00:48:45.670 Brian McGough: do

584 00:48:46.730 00:48:51.460 Brian McGough: now to like? I I guess what we want is a demo right.

585 00:48:51.460 00:48:52.020 Uttam Kumaran: Yeah.

586 00:48:52.830 00:48:54.849 Brian McGough: So what do we want to demo.

587 00:48:57.110 00:49:02.610 Uttam Kumaran: So I think the couple of things that would be great to Demo is one having.

588 00:49:03.150 00:49:04.070 Uttam Kumaran: just like.

589 00:49:04.930 00:49:08.910 Uttam Kumaran: if you have a repository of Pdfs of market data

590 00:49:08.950 00:49:11.000 Uttam Kumaran: and other like

591 00:49:11.210 00:49:19.579 Uttam Kumaran: things on the on the landlord side. Can we ask questions over those structured documents where it extracts real information out of it.

592 00:49:19.890 00:49:22.539 Uttam Kumaran: I think that’s the initial demo.

593 00:49:22.600 00:49:26.460 Uttam Kumaran: I think we can then probably leverage some of that stuff for

594 00:49:26.540 00:49:30.030 Uttam Kumaran: the tenant side, which is like extracting data from the market rates.

595 00:49:30.470 00:49:35.469 Brian McGough: Do you have any? Do you have like those like any Pdf examples.

596 00:49:36.410 00:49:42.879 Uttam Kumaran: I do. Yeah. So this is pro. This is the. This is like a Pdf of an investment memo about

597 00:49:43.230 00:49:45.109 Uttam Kumaran: like an opportunity

598 00:49:45.130 00:49:47.950 Uttam Kumaran: where there’s a ton of structured data everywhere.

599 00:49:48.000 00:49:53.919 Uttam Kumaran: And basically, maybe our first goal is to figure out how to chunk and parse.

600 00:49:54.300 00:49:57.500 Uttam Kumaran: And like, have this data ready for.

601 00:49:57.950 00:49:58.790 Brian McGough: Yeah.

602 00:49:59.180 00:50:00.480 Brian McGough: okay, let’s

603 00:50:01.920 00:50:04.569 Brian McGough: Let’s fucking. Try to do this in Agent Hub.

604 00:50:06.390 00:50:08.439 Brian McGough: have you tried to do that at all? There.

605 00:50:08.440 00:50:10.220 Uttam Kumaran: No, no, I’ve not tried anything.

606 00:50:10.390 00:50:16.222 Brian McGough: Okay. Hell, yeah, let’s, this is this is not. This is what I’m looking for. Alright cool. Let’s do that. Let’s

607 00:50:18.340 00:50:19.930 Brian McGough: Maybe.

608 00:50:20.690 00:50:23.040 Uttam Kumaran: So basically, what you think we would do is like.

609 00:50:24.000 00:50:26.480 Uttam Kumaran: have a file reader or something.

610 00:50:26.630 00:50:28.909 Uttam Kumaran: Or maybe it reads from the Google Drive.

611 00:50:29.200 00:50:30.080 Uttam Kumaran: And then.

612 00:50:30.720 00:50:32.749 Brian McGough: Do we have to create like a new

613 00:50:33.530 00:50:35.749 Brian McGough: flow? Or is this this should be.

614 00:50:35.750 00:50:38.090 Uttam Kumaran: I would. Yeah, I’ll just create a new one.

615 00:50:47.802 00:50:49.490 Uttam Kumaran: Go ahead and rename.

616 00:50:50.100 00:50:51.240 Uttam Kumaran: So

617 00:50:51.650 00:50:54.656 Uttam Kumaran: maybe we could just think about how we would do this.

618 00:50:56.340 00:50:56.970 Brian McGough: Yeah.

619 00:51:00.140 00:51:05.409 Brian McGough: there’s definitely I’m I’m curious about the data loaders. Like I, I feel like there’s a Pdf thing.

620 00:51:05.690 00:51:07.590 Uttam Kumaran: There is a file. Reader.

621 00:51:07.590 00:51:08.930 Brian McGough: You scroll down a little bit.

622 00:51:11.680 00:51:12.330 Uttam Kumaran: Do you get? But yeah.

623 00:51:12.650 00:51:13.290 Brian McGough: Pedia freeer.

624 00:51:14.470 00:51:15.699 Brian McGough: Yeah, let’s try that.

625 00:51:18.050 00:51:20.769 Brian McGough: Let’s try that. We can try like a few different ones.

626 00:51:25.230 00:51:27.240 Brian McGough: So I think we need to.

627 00:51:29.880 00:51:31.690 Brian McGough: Yeah, okay, so

628 00:51:34.040 00:51:35.849 Brian McGough: do do, do, do.

629 00:51:36.090 00:51:37.729 Uttam Kumaran: Maybe just ask AI.

630 00:51:41.970 00:51:42.760 Brian McGough: Yeah.

631 00:51:43.390 00:51:45.520 Uttam Kumaran: Let’s just try. Let’s just see what happens.

632 00:51:47.560 00:51:48.929 Brian McGough: Like, can we

633 00:51:49.240 00:51:54.120 Brian McGough: take that? Those Pdf contents and type that in as a context.

634 00:51:54.120 00:51:54.780 Uttam Kumaran: Yeah.

635 00:51:55.570 00:51:56.370 Uttam Kumaran: Oh.

636 00:51:58.390 00:52:00.680 Brian McGough: List of text is not equal to text.

637 00:52:00.790 00:52:05.749 Brian McGough: So maybe it’s in, maybe in the data loaders, we need to like junk or something

638 00:52:06.110 00:52:07.840 Brian McGough: extract item list.

639 00:52:11.030 00:52:14.769 Brian McGough: Well, this is this is this is what the opposite of what we need. Kind of.

640 00:52:15.170 00:52:16.780 Brian McGough: Oh, combine lists.

641 00:52:16.810 00:52:18.869 Brian McGough: Oh, yeah, data modifiers. Maybe.

642 00:52:20.780 00:52:21.860 Brian McGough: No.

643 00:52:22.820 00:52:24.689 Brian McGough: we need to flatten

644 00:52:26.130 00:52:28.150 Brian McGough: join list items. Maybe.

645 00:52:30.910 00:52:35.439 Brian McGough: I guess the point is like, what we need to do is we need, yeah, this. Pdf.

646 00:52:37.240 00:52:39.490 Brian McGough: yeah, I mean, I guess we yeah, we could try this.

647 00:52:39.680 00:52:42.219 Uttam Kumaran: Let’s just try it. And then but I honestly stick.

648 00:52:43.030 00:52:45.540 Uttam Kumaran: We should just do this in like a notebook.

649 00:52:46.710 00:52:47.450 Uttam Kumaran: Yeah.

650 00:52:48.270 00:52:48.980 Brian McGough: That’s.

651 00:52:49.300 00:52:52.820 Uttam Kumaran: Because I saw some of those Demos, and they looked like, I think it’s

652 00:52:53.640 00:52:54.980 Uttam Kumaran: let’s just try this.

653 00:52:57.600 00:53:00.979 Brian McGough: We need to somehow add this document.

654 00:53:01.290 00:53:03.100 Uttam Kumaran: Yeah, let me just download it.

655 00:53:07.990 00:53:08.630 Uttam Kumaran: Die.

656 00:53:08.630 00:53:12.490 Brian McGough: Yeah, how do you? I think you need to like, put it in Agent Hub.

657 00:53:16.760 00:53:17.719 Uttam Kumaran: Oh, there you go!

658 00:53:17.870 00:53:20.329 Brian McGough: Top left here. Maybe there’s this little folder.

659 00:53:21.690 00:53:22.480 Uttam Kumaran: Up.

660 00:53:41.510 00:53:42.230 Brian McGough: Sick.

661 00:54:05.650 00:54:06.690 Brian McGough: nice.

662 00:54:08.410 00:54:10.879 Uttam Kumaran: Okay, so let’s think about a couple of other

663 00:54:11.470 00:54:13.680 Uttam Kumaran: things. We can ask it.

664 00:54:14.455 00:54:14.800 Uttam Kumaran: Feeling.

665 00:54:20.610 00:54:21.420 Brian McGough: Fancy.

666 00:54:22.670 00:54:25.590 Brian McGough: Oh, shit! This is with our account. I never do this.

667 00:54:26.390 00:54:35.137 Uttam Kumaran: This is arc. I like arc. It’s not like as crazy as like some of the Fan boys kind of described this kind of way, but it’s fine.

668 00:54:35.450 00:54:36.330 Brian McGough: I agree.

669 00:54:36.700 00:54:37.750 Brian McGough: Pandora

670 00:54:37.890 00:54:42.669 Brian McGough: sold it to me as like the be all end, all of the Internet. And I was like, alright.

671 00:54:42.870 00:54:44.383 Uttam Kumaran: Yeah, I.

672 00:54:45.140 00:54:49.009 Brian McGough: What’s the invest equity investment? Multiple? Ask that question.

673 00:54:49.010 00:54:52.049 Uttam Kumaran: Where is that? Oh, okay, okay.

674 00:55:23.980 00:55:25.050 Uttam Kumaran: Yikes.

675 00:55:25.990 00:55:27.400 Uttam Kumaran: what? The fuck?

676 00:55:29.450 00:55:30.440 Uttam Kumaran: Oh.

677 00:55:30.760 00:55:32.069 Uttam Kumaran: this is the target.

678 00:55:34.420 00:55:35.740 Uttam Kumaran: It’s projective.

679 00:55:36.470 00:55:41.820 Uttam Kumaran: Okay? So see that there’s a there’s an issue right like this is like about their track record.

680 00:55:42.480 00:55:43.080 Brian McGough: Yup!

681 00:55:43.250 00:55:44.490 Uttam Kumaran: But you’re right

682 00:55:44.640 00:55:48.059 Uttam Kumaran: so. But can I get it to tell me where the fucking came from?

683 00:55:50.052 00:55:52.308 Brian McGough: So there are.

684 00:55:53.300 00:55:59.209 Brian McGough: There are ways to do this with some lank chain stuff that I’ve been or

685 00:55:59.830 00:56:03.610 Brian McGough: sit, length, change, and index. I get the 2 so confused.

686 00:56:04.089 00:56:10.470 Brian McGough: Either with lang chain or long index. I I have to like look through, but I saw something recently where

687 00:56:12.600 00:56:17.210 Brian McGough: where it provides. Like we’re we’re like the document that

688 00:56:20.320 00:56:28.779 Brian McGough: contains, the information that is referenced will like you like, we basically just like will display sources.

689 00:56:29.160 00:56:30.040 Uttam Kumaran: Hmm.

690 00:56:30.570 00:56:31.730 Uttam Kumaran: okay. Okay.

691 00:56:31.890 00:56:34.570 Brian McGough: I really like sources, because I just think

692 00:56:34.610 00:56:35.879 Brian McGough: you know, you need to.

693 00:56:35.880 00:56:41.250 Uttam Kumaran: No, I I only want to do. I don’t want it to. I want us to have a really concrete understanding of it’s hallucinating.

694 00:56:42.330 00:56:43.055 Uttam Kumaran: Yep.

695 00:56:46.450 00:56:47.439 Brian McGough: But so here.

696 00:56:47.440 00:56:49.240 Uttam Kumaran: Here’s an example of like

697 00:56:52.460 00:56:56.232 Uttam Kumaran: what some people were doing in like a Pdf parsing.

698 00:56:58.160 00:56:59.340 Brian McGough: Yeah. Lala parse.

699 00:57:00.520 00:57:01.290 Uttam Kumaran: Yeah. Just

700 00:57:01.630 00:57:02.676 Uttam Kumaran: Api key.

701 00:57:03.350 00:57:05.009 Uttam Kumaran: like, hold on to.

702 00:57:05.010 00:57:06.240 Brian McGough: Or can you add this in the chat.

703 00:57:06.582 00:57:14.450 Uttam Kumaran: Close attention to you know what you’re saving these as right. So this is the file that you’ll need to

704 00:57:15.045 00:57:15.520 Uttam Kumaran: spend

705 00:57:16.430 00:57:18.520 Uttam Kumaran: query engine. To see if this is

706 00:57:18.710 00:57:20.640 Uttam Kumaran: is just in met

707 00:57:21.090 00:57:23.840 Uttam Kumaran: is true of this retrieval process.

708 00:57:24.964 00:57:35.769 Uttam Kumaran: What is the growth carrying amount of total, amortizable, intangible assets for January 20, ninth formation. That would be wrong. Right? So the the the

709 00:57:39.925 00:57:41.410 Uttam Kumaran: let me send this to you in slack.

710 00:57:42.570 00:57:43.220 Brian McGough: Thank you.

711 00:57:46.930 00:57:50.219 Brian McGough: See, I’ve just been watching so many

712 00:57:50.250 00:57:51.300 Brian McGough: videos.

713 00:57:51.300 00:57:51.980 Uttam Kumaran: So that.

714 00:57:55.090 00:57:56.340 Brian McGough: Which is.

715 00:57:57.570 00:57:59.730 Brian McGough: I guess, normal for me. But.

716 00:58:00.070 00:58:02.750 Uttam Kumaran: No, I mean I did for like 3 months.

717 00:58:03.120 00:58:03.939 Brian McGough: What’s that?

718 00:58:04.900 00:58:06.629 Uttam Kumaran: That’s what I did for like 3 months.

719 00:58:07.080 00:58:09.269 Brian McGough: Yeah, they just click, slide.

720 00:58:09.510 00:58:10.999 Brian McGough: Yeah, dude, like.

721 00:58:11.160 00:58:13.219 Brian McGough: just so many

722 00:58:16.720 00:58:20.889 Brian McGough: but learning a lot of a ton of dude. I listen to this fucking

723 00:58:21.120 00:58:23.650 Brian McGough: podcasts last night with

724 00:58:24.650 00:58:26.080 Brian McGough: with a

725 00:58:26.150 00:58:30.740 Brian McGough: I’ve I’ve been watching a lot of Jerry Lou, shit dude. That guy’s on point.

726 00:58:31.570 00:58:32.369 Uttam Kumaran: Who is that?

727 00:58:33.090 00:58:34.538 Brian McGough: Jerry Lou is the

728 00:58:34.910 00:58:36.649 Brian McGough: CEO of Lom index.

729 00:58:37.030 00:58:39.079 Uttam Kumaran: Oh, yeah. Yeah, yeah. I know. Okay, okay.

730 00:58:39.662 00:58:40.870 Brian McGough: Found this fucking

731 00:58:41.938 00:58:47.860 Brian McGough: like production ready. They have a they have a I I sent it in in slack, but.

732 00:58:47.860 00:58:49.560 Uttam Kumaran: Moment, I know. But this guy.

733 00:58:50.580 00:58:52.035 Brian McGough: Yeah, that’s him.

734 00:58:53.200 00:58:53.930 Brian McGough: he

735 00:58:57.570 00:58:58.740 Brian McGough: So

736 00:59:01.934 00:59:08.129 Brian McGough: they have a they have like a repo. Or they made. They made this like website that

737 00:59:08.800 00:59:10.420 Brian McGough: analyzes

738 00:59:10.700 00:59:13.779 Brian McGough: Dot and compares like financial documents.

739 00:59:14.420 00:59:16.060 Brian McGough: And they basically just like

740 00:59:16.320 00:59:19.990 Brian McGough: made open source. The repo.

741 00:59:21.500 00:59:22.410 Uttam Kumaran: Just like a.

742 00:59:22.800 00:59:24.840 Brian McGough: Full stack, production, ready.

743 00:59:25.610 00:59:26.700 Brian McGough: rag.

744 00:59:27.030 00:59:28.210 Brian McGough: application.

745 00:59:28.210 00:59:29.359 Uttam Kumaran: Oh, nice!

746 00:59:29.620 00:59:30.460 Brian McGough: Copy.

747 00:59:32.520 00:59:33.809 Uttam Kumaran: I’m sick. Okay.

748 00:59:34.190 00:59:34.970 Uttam Kumaran: yeah.

749 00:59:39.080 00:59:45.289 Uttam Kumaran: okay. So I wonder if you like you, you want to just play around with this stuff today. And then we we can catch up.

750 00:59:46.050 00:59:48.631 Uttam Kumaran: I’m gonna be driving this afternoon, but

751 00:59:49.480 00:59:50.789 Brian McGough: Oh, yeah, you’re going to Dallas.

752 00:59:51.100 00:59:53.720 Uttam Kumaran: Yeah, I’m going to Dallas with like 3 Ish.

753 00:59:54.050 00:59:56.420 Uttam Kumaran: But I’m gonna be online tomorrow.

754 00:59:56.630 01:00:03.669 Uttam Kumaran: But basically like, how do you like, maybe I’m just gonna throw this in the Google drive. And do, if you use Google Collab, you can hook up to drive pretty easily.

755 01:00:03.850 01:00:04.580 Uttam Kumaran: Yeah.

756 01:00:04.700 01:00:06.490 Uttam Kumaran: you want to just start a notebook.

757 01:00:06.590 01:00:08.900 Uttam Kumaran: and then we can like pair in there.

758 01:00:10.152 01:00:11.850 Brian McGough: Sure. Yeah, I mean, I think

759 01:00:12.380 01:00:21.040 Brian McGough: I think I think this is an interesting problem right here that we’re looking at right. And this is, I feel like, gonna be where a lot of the

760 01:00:21.480 01:00:24.160 Brian McGough: engineering work goes into

761 01:00:24.190 01:00:25.270 Brian McGough: is like.

762 01:00:26.030 01:00:29.340 Brian McGough: how do we process this data? And how do we

763 01:00:29.920 01:00:33.070 Brian McGough: like, try to trick it, and try to

764 01:00:33.860 01:00:40.689 Brian McGough: make it hallucinate, or give wrong answers, and have, like 2 similar pieces of information in a

765 01:00:40.930 01:00:43.030 Brian McGough: Pdf. Or what have you?

766 01:00:43.080 01:00:45.239 Brian McGough: And then, how do we?

767 01:00:46.670 01:00:48.690 Brian McGough: Yeah, how do we like.

768 01:00:51.870 01:00:56.540 Brian McGough: you know, distinguish between these disparate pieces of information? How do we?

769 01:00:58.860 01:01:03.230 Brian McGough: Yeah, it’s it’s a very. It’s an interesting question. Lot of interesting questions.

770 01:01:03.450 01:01:10.220 Uttam Kumaran: Yeah, I think I think at minimum, probably pretty quickly, we can get to the point where we ask questions over it. I think.

771 01:01:10.560 01:01:16.459 Uttam Kumaran: yeah, you’re right, like, what we’re gonna have to figure out is, can it say, oh, we got this answer from page 7.

772 01:01:17.430 01:01:18.690 Uttam Kumaran: Or.

773 01:01:19.090 01:01:22.259 Uttam Kumaran: for example, in this document. There’s

774 01:01:22.270 01:01:24.229 Uttam Kumaran: like structured tables.

775 01:01:24.330 01:01:29.219 Uttam Kumaran: Which is why there’s point in that one video where it’s like, how do you extract shit like this

776 01:01:30.400 01:01:33.660 Uttam Kumaran: like dude? If you were to just scrape this as text. It comes out like

777 01:01:34.250 01:01:35.940 Uttam Kumaran: it’s like all messed up.

778 01:01:36.150 01:01:38.460 Uttam Kumaran: So you almost need to retain the structure.

779 01:01:39.970 01:01:41.439 Brian McGough: Yeah, I was just watching video.

780 01:01:41.440 01:01:42.070 Uttam Kumaran: Brian.

781 01:01:44.410 01:01:47.189 Brian McGough: Just watching something where they talk about this.

782 01:01:48.300 01:01:57.720 Uttam Kumaran: I think it. I mean, yeah, there’s a lot like I. This is what I was sending you about. The Ocr thing which is like this is why sometimes Ocr is matters because

783 01:01:57.890 01:02:03.570 Uttam Kumaran: you, if you were to scrape this as text, you don’t get this structure. But if you were to take a screenshot of it.

784 01:02:04.150 01:02:07.809 Uttam Kumaran: you actually see that? Oh, it is. It is almost like a tabular format.

785 01:02:08.760 01:02:12.950 Uttam Kumaran: So that’s why our, our, the Pdf processing process, I think, is gonna be

786 01:02:13.430 01:02:15.959 Uttam Kumaran: like way more complicated than the

787 01:02:18.470 01:02:22.520 Uttam Kumaran: like the the queue the question to answer, because everything just gets put in this text.

788 01:02:24.270 01:02:25.909 Uttam Kumaran: The other thing that’s

789 01:02:25.940 01:02:27.429 Uttam Kumaran: interesting is like

790 01:02:27.600 01:02:44.490 Uttam Kumaran: these are, we may be able to make some assumptions about the this document that help us. For example, we can say, this document contains Xyz. It’s formatted in this way. There are these sorts of elements in here. Like, I think it’s actually helpful to know that this is like the structure of a

791 01:02:44.590 01:02:46.532 Uttam Kumaran: of a document.

792 01:02:47.320 01:02:52.190 Uttam Kumaran: but like, for example, this is like a really great thing where you would want to ask, what’s the legal accounting

793 01:02:52.690 01:02:54.130 Uttam Kumaran: like cost

794 01:02:54.720 01:02:55.570 Uttam Kumaran: for the

795 01:02:55.710 01:02:56.820 Uttam Kumaran: last minute?

796 01:02:58.900 01:02:59.720 Uttam Kumaran: You know.

797 01:03:00.070 01:03:00.660 Brian McGough: Yep.

798 01:03:02.260 01:03:02.970 Brian McGough: so

799 01:03:04.570 01:03:05.590 Brian McGough: I think

800 01:03:05.810 01:03:13.339 Brian McGough: it sounds like what I could just hack on for for the time being. Is

801 01:03:15.930 01:03:19.169 Brian McGough: like answering questions over this document.

802 01:03:22.220 01:03:22.970 Uttam Kumaran: right?

803 01:03:23.910 01:03:26.119 Uttam Kumaran: Yeah, let’s just do that. And I think.

804 01:03:27.075 01:03:30.000 Uttam Kumaran: let’s just do that for a week, I mean fuck it.

805 01:03:30.190 01:03:32.660 Brian McGough: Yeah, I think that’s probably like

806 01:03:32.950 01:03:34.230 Brian McGough: time well spent.

807 01:03:34.230 01:03:38.690 Uttam Kumaran: I think the PI think again we’ll learn a lot about Pdf parsing and the process

808 01:03:39.190 01:03:41.449 Uttam Kumaran: that will go directly to

809 01:03:41.550 01:03:42.870 Uttam Kumaran: the tenant side.

810 01:03:42.940 01:03:51.085 Uttam Kumaran: And that’s frankly what I mean, honestly like depending on how far we get. We could even try. We could just like shove market other market stuff in there and ask questions.

811 01:03:51.450 01:04:00.860 Uttam Kumaran: And basically, like, I think that’s probably good enough for a demo. The other thing that they are asking about is the ability to generate

812 01:04:01.480 01:04:02.530 Uttam Kumaran: charts

813 01:04:02.550 01:04:06.740 Uttam Kumaran: on the on the question. Answer side. So the one thing I I shared was like.

814 01:04:06.890 01:04:09.490 Uttam Kumaran: There’s this Vercell AI.

815 01:04:09.490 01:04:12.762 Brian McGough: So, so like, so like you have the 3 different

816 01:04:13.200 01:04:17.540 Brian McGough: comparable scenarios, and you want to like chart them, or something.

817 01:04:19.450 01:04:23.470 Uttam Kumaran: Yeah. But basically, it’s like, can you generate the chart itself?

818 01:04:24.130 01:04:25.000 Uttam Kumaran: Right?

819 01:04:25.380 01:04:26.109 Uttam Kumaran: I mean.

820 01:04:26.110 01:04:26.640 Brian McGough: Open! AI!

821 01:04:26.640 01:04:27.270 Uttam Kumaran: Why, they fuckin’.

822 01:04:27.270 01:04:28.499 Brian McGough: Police can do that.

823 01:04:32.020 01:04:33.620 Uttam Kumaran: Let me show you

824 01:04:35.280 01:04:36.210 Uttam Kumaran: this.

825 01:04:43.330 01:04:45.520 Uttam Kumaran: Can you hear the audio from these? By the way.

826 01:04:47.270 01:04:48.470 Brian McGough: Let’s see.

827 01:04:52.240 01:04:59.120 Uttam Kumaran: AI. We’ve been working on this with our product called where you can give it some prompts like an e-commerce dashboard. With this.

828 01:04:59.120 01:05:00.999 Brian McGough: Oh, yeah, this is yourself, you know.

829 01:05:01.710 01:05:07.959 Uttam Kumaran: Or you can upload an image for you, so you can see we’ve got this dashboard app.

830 01:05:08.140 01:05:35.719 Uttam Kumaran: I can go to Code, and I can run this in the Cli, or I can copy paste this code into my application. Up until now this experience has been built in as part of the product. So the underlying code is not open source. But we wanted to make this technology available for everyone and open source. So we’re excited to share. We’ve released a new version of the AI SDK that allows you to do just that. So we have the demo to show some of the possibilities which can build with this technology that we’re calling generated live. So let’s take a look if you want to. First of all.

831 01:05:36.030 01:05:50.739 Uttam Kumaran: you you have some call here for whether to adjust on my representation of what the Lm needs. So maybe it’s like your textual conversation in the chat app. And then the Ui state react. This helps keep the data in the Ui elements that are turned by Lm insane.

832 01:05:50.740 01:06:14.760 Uttam Kumaran: So check us out and we’ll learn more about how to build with the A. SDK. I’m gonna walk through a bunch of other really awesome examples from the community, and we’ll put some links down below to it’s like I started building with the isdk. This example is another nice e-commerce example, showing a list of product cards, and then also showing how you could embed something like a one time password code. They could send to your phone, or you can enter in the numbers here. These next 2 are interesting ways. You could think about visualizing data, either from a database or using rag. And in this example, we’re looking up analytics data. So coming from post talk here.

833 01:06:14.900 01:06:38.409 Uttam Kumaran: I wanna ask, what would my page use between some specific date? And then I could also ask it to sum up the count. So this is also showing a list of the specific pages and underlying sequel that’s generating another example here, doing something pretty similar, but doing rag reading from a superb base post press database again showing a chart at an actual sequel that’s being ran very cool. The next view was for the idea of interactive quizzes. So actually, you can actually generate the sequel, and then also generate the chart ui with the data

834 01:06:38.810 01:06:40.060 Uttam Kumaran: like on the fly.

835 01:06:40.573 01:06:44.449 Uttam Kumaran: I think that’s like kind of interesting on the front end side.

836 01:06:44.460 01:06:53.770 Uttam Kumaran: For example, it’s like, can you based on the type of response? Can you structure what the user sees, which is either a chart or a string answer or a.

837 01:06:53.770 01:06:54.490 Brian McGough: Mix.

838 01:06:54.930 01:06:57.369 Uttam Kumaran: So that could be interesting to play with.

839 01:06:57.400 01:06:58.470 Uttam Kumaran: It’s probably

840 01:06:59.360 01:07:02.200 Uttam Kumaran: yeah, probably, like the 2 things we can take a look at

841 01:07:02.640 01:07:03.440 Uttam Kumaran: cool.

842 01:07:03.866 01:07:05.573 Brian McGough: Alright. So I will.

843 01:07:06.280 01:07:12.370 Brian McGough: I have that I have in the email that you’ve shared with me.

844 01:07:13.320 01:07:18.150 Brian McGough: Do I have that document? Yes, wait, Nope, I don’t.

845 01:07:18.924 01:07:21.120 Uttam Kumaran: I’m just gonna share you on this Google drive

846 01:07:21.540 01:07:22.230 Uttam Kumaran: cool.

847 01:07:24.760 01:07:29.160 Brian McGough: sweet, and then I will just try to rag that shit up.

848 01:07:29.970 01:07:31.420 Uttam Kumaran: Yeah.

849 01:07:33.450 01:07:34.130 Uttam Kumaran: thank, you.

850 01:07:34.130 01:07:35.970 Brian McGough: Greg. Is life right now?

851 01:07:36.422 01:07:39.587 Uttam Kumaran: Yeah, I’m gonna name my first child.

852 01:07:40.040 01:07:43.569 Brian McGough: Hag Dude. That’s a good name for a boy or a girl.

853 01:07:43.830 01:07:45.040 Uttam Kumaran: Greg.

854 01:07:45.875 01:07:46.660 Uttam Kumaran: right.

855 01:07:46.790 01:07:49.040 Uttam Kumaran: Yo big r.

856 01:07:49.410 01:07:50.319 Brian McGough: Big off

857 01:07:53.220 01:07:59.530 Brian McGough: dude alright cool. Let me make sure I got this file. Sweep. Got it

858 01:08:00.660 01:08:01.610 Brian McGough: alright, man!

859 01:08:02.100 01:08:06.330 Uttam Kumaran: Yeah. And like, may I? Again, I think I think if you want to do everything in typescript.

860 01:08:06.400 01:08:09.959 Uttam Kumaran: feel free, or I think a lot of the examples are in Collab, whatever it works.

861 01:08:09.960 01:08:11.909 Brian McGough: Dude. I know I’m I’m like.

862 01:08:12.150 01:08:13.395 Uttam Kumaran: Whatever works, cause.

863 01:08:13.810 01:08:20.850 Brian McGough: Interesting using python. But the more I like getting these different libraries, the more you know the python libraries.

864 01:08:20.859 01:08:23.500 Brian McGough: Just see more robust. And, like.

865 01:08:24.300 01:08:28.131 Brian McGough: you know, there’s, you know, like I I went to use

866 01:08:29.600 01:08:33.419 Brian McGough: what was I trying to use? I was trying to use some tool

867 01:08:35.134 01:08:41.129 Brian McGough: yesterday, and like I was, I’ve been using Lang Smith and the node SDK. And then.

868 01:08:41.670 01:08:47.610 Brian McGough: like, I went to use a tool like another Lang Smith tool, and it was just like not in

869 01:08:47.680 01:08:48.870 Brian McGough: node at all.

870 01:08:48.910 01:08:50.540 Brian McGough: So I was like fuck.

871 01:08:51.160 01:08:54.800 Brian McGough: and and they’re like, alright. I have to build a separate service to like.

872 01:08:55.680 01:09:03.469 Brian McGough: you know, with python, so I might just I might just try to do shit in python. I don’t know, especially with Co. Lab, you know, like, have.

873 01:09:03.479 01:09:06.179 Uttam Kumaran: And that’s the thing. The call out stuff is so easy.

874 01:09:06.470 01:09:07.779 Brian McGough: It’s so nice.

875 01:09:09.180 01:09:09.710 Uttam Kumaran: So.

876 01:09:09.710 01:09:14.180 Brian McGough: But cool. Yeah, I’ll fuck around with this. Maybe I’ll just try to do this in Co lab.

877 01:09:14.590 01:09:15.250 Uttam Kumaran: Okay.

878 01:09:16.580 01:09:17.370 Brian McGough: Alright, bro.

879 01:09:17.720 01:09:19.260 Uttam Kumaran: I did talk to you in slack.

880 01:09:19.520 01:09:20.680 Brian McGough: Alright, man, good talk later.

881 01:09:20.689 01:09:21.569 Uttam Kumaran: Okay. Please.