Meeting Title: Brainforge x Ujval Kamath Collaboration Discussion Date: 2026-01-15 Meeting participants: Ujval Kamath, Uttam Kumaran
WEBVTT
1 00:16:14.130 ⇒ 00:16:17.810 Uttam Kumaran: Hello! So sorry for the delay. Just,
2 00:16:18.510 ⇒ 00:16:21.720 Uttam Kumaran: It’s, first, like, 2 weeks back from…
3 00:16:21.870 ⇒ 00:16:32.729 Uttam Kumaran: break, and I… yeah, we’re just growing, and there’s a lot of… it’s all good stuff, but I appreciate your time. I did not mean to… to be… to be late today, so thank you.
4 00:16:33.590 ⇒ 00:16:37.620 Ujval Kamath: Oh, it’s not a problem. I’ve been in customer-facing roles, I know how it goes, so…
5 00:16:37.970 ⇒ 00:16:56.879 Uttam Kumaran: No, I know, I, I, I appreciate you taking the time and even, you know, spending time work, you know, talking to me, so, it’s good, though. We’re, we’re pitching, you know, we do a lot of product analytics work, which I’m sure you’re… you’re very familiar with, and so we… we’re… we… one of our clients is a… is like a…
6 00:16:57.180 ⇒ 00:17:05.669 Uttam Kumaran: an AI video generating, like, they’re a startup, and they’re growing a lot, and they’re doing a lot of product
7 00:17:05.960 ⇒ 00:17:08.789 Uttam Kumaran: Prioritization based on, kind of, gut feeling.
8 00:17:08.839 ⇒ 00:17:28.719 Uttam Kumaran: And they’re losing a lot of users, and we’re like, you need some type of product analytics strategy here, because you’re going to pay for all these folks, and they’re gonna get kicked right out the door, and so that was just our… the conversation I came from, just trying to share with them, like, hey, well, I think we can help you guys implement a strategy there, so…
9 00:17:29.250 ⇒ 00:17:29.960 Ujval Kamath: Okay.
10 00:17:31.600 ⇒ 00:17:34.550 Uttam Kumaran: Awesome. Well, yeah, how do you know Jody, by the way?
11 00:17:35.020 ⇒ 00:17:39.559 Ujval Kamath: So, I… do you know who Joe Rye… Joe… I can never say his last name.
12 00:17:39.560 ⇒ 00:17:41.539 Uttam Kumaran: Reese? Yeah, Reese.
13 00:17:41.540 ⇒ 00:17:49.660 Ujval Kamath: Yeah, I can’t… I don’t know how to pronounce his last name, but he has, like, an online forum, which is primarily focused on data.
14 00:17:49.660 ⇒ 00:17:50.930 Uttam Kumaran: I’m in there, yeah, I’m in there.
15 00:17:50.930 ⇒ 00:17:51.960 Ujval Kamath: Oh, yeah.
16 00:17:51.960 ⇒ 00:17:55.680 Uttam Kumaran: all the time, but yeah, I, I… Okay. It’s a great, it’s a great group.
17 00:17:55.860 ⇒ 00:17:56.490 Ujval Kamath: Yeah.
18 00:17:56.720 ⇒ 00:18:01.130 Uttam Kumaran: So I, I just, I, I, I post sometimes on there, more so on the…
19 00:18:01.130 ⇒ 00:18:07.230 Ujval Kamath: the sort of data science and AI stuff, I don’t post that much of the engine, but that’s how I met him, and I met a couple of other people.
20 00:18:07.340 ⇒ 00:18:12.629 Ujval Kamath: So… And, yep, that’s… it sort of just went on from there.
21 00:18:12.890 ⇒ 00:18:13.630 Uttam Kumaran: Nice
22 00:18:13.760 ⇒ 00:18:28.900 Uttam Kumaran: Yeah, I met Jody through a mutual friend, who I did some work with, like, years ago, and then she was like, hey, Jody’s also in, like, data consulting, like, you guys should chat. So we became friends, and yeah, he’s helped out our company a bunch, so…
23 00:18:29.500 ⇒ 00:18:30.130 Ujval Kamath: Okay.
24 00:18:31.010 ⇒ 00:18:39.789 Uttam Kumaran: Yeah, well, it’s really great to meet. I, would love to, you know, tell you a little bit about Brainforge and, you know, hear a little bit about your background and
25 00:18:39.790 ⇒ 00:18:56.570 Uttam Kumaran: sort of see… tell you a little bit, sort of, opportunities and kind of where our company’s going, and… and hear a bit about, like, where… what you’re interested in doing, see if maybe there’s a fit. So… so Brainforge is a data and AI consultancy. It’s a business I started a few years ago. My background is in computer engineering and data engineering.
26 00:18:56.570 ⇒ 00:19:03.140 Uttam Kumaran: worked in New York for a few years at multiple, you know, kind of increasingly smaller startups.
27 00:19:03.350 ⇒ 00:19:12.379 Uttam Kumaran: Doing data work, leading data teams, and then most previously led product as a data company. And then in between there, I had done some consulting.
28 00:19:12.430 ⇒ 00:19:24.550 Uttam Kumaran: And sort of learned a little bit about what it’s like to sort of just be a contract engineer on projects. And then decided to sort of start this business, a few years ago, and sort of have just been kind of growing ever since, really.
29 00:19:24.900 ⇒ 00:19:28.659 Uttam Kumaran: Kind of, like, bootstrapped the entire thing. So, our company is now…
30 00:19:28.680 ⇒ 00:19:32.930 Uttam Kumaran: Around, like, 20, 25 people, mostly data.
31 00:19:32.930 ⇒ 00:19:56.499 Uttam Kumaran: data, you know, and full-stack engineers that are doing AI work. They’re not AI engineers, I guess, but, you know, a lot of, like, back-end, full-stack folks. So we really, a lot of our work is around helping, you know, companies just make more decisions and, make more accurate decisions. And so, so really everything around leveraging data and AI for what we’re calling, like, sort of decision intelligence.
32 00:19:56.710 ⇒ 00:20:11.699 Uttam Kumaran: So for us, we sort of do everything from data warehousing, ETL, data modeling, all the way through BI, as well as, as I mentioned, like, product analytics. But we also do a lot of strategy, so strategy recommendations, more like,
33 00:20:11.900 ⇒ 00:20:22.830 Uttam Kumaran: sort of the types of analysis you would expect out of, like, you know, a Bain or a BCG, where we’re kind of canvassing, a problem and sort of producing a lot of insight around it.
34 00:20:22.880 ⇒ 00:20:38.709 Uttam Kumaran: And so we… it’s been awesome. We have an amazing, amazing team of great data and AI folks. We really… again, our bread and butter is data, but I built the business in the last 3 years heavily using AI for internal workflows.
35 00:20:38.710 ⇒ 00:20:43.340 Uttam Kumaran: Basically, across the entirety of the business, from sales, finance,
36 00:20:43.340 ⇒ 00:21:00.929 Uttam Kumaran: operations, delivery, of course, on the engineering side. And so, we really learned firsthand what it was like to build, you know, whether it’s AI workflows, build agents, sort of work with all these, you know, new frameworks.
37 00:21:00.950 ⇒ 00:21:05.870 Uttam Kumaran: And so we decided, you know, about a year and a half ago to also start going to market with services.
38 00:21:06.020 ⇒ 00:21:07.520 Uttam Kumaran: Around, AI.
39 00:21:07.560 ⇒ 00:21:22.369 Uttam Kumaran: And so, yeah, that’s kind of, like, how we’ve sort of grown the business so far. We typically work with companies anywhere from 20 million in annual revenue, and we have a couple people in the portfolio that are a few hundred million in revenue, and it’s sort of, like, growing,
40 00:21:22.370 ⇒ 00:21:36.329 Uttam Kumaran: From there, we’ve done a lot of work in CPG, omnichannel retail, e-commerce, as well as in B2B and B2C SaaS, as well as, you know, some one-offs in, you know, health and legal and insurance.
41 00:21:37.790 ⇒ 00:21:47.100 Uttam Kumaran: But yeah, that’s sort of a little bit about the story of us, and yeah, I’d love to kind of hear about your background, and sort of even, like, what interested you in,
42 00:21:47.230 ⇒ 00:21:49.980 Uttam Kumaran: And, like, chatting with… chatting with me and chatting about us.
43 00:21:51.240 ⇒ 00:22:04.340 Ujval Kamath: Oh, sure. So, maybe, if I talk about it myself a little bit, it might explain why I… I found… I saw it when Jody told me about Brainforges. I said, oh, this… this maybe is a little… is kind of consistent with my background, with at least what I’m interested in.
44 00:22:04.340 ⇒ 00:22:13.190 Ujval Kamath: So, yeah, so I’m Oujwal, I’ve been a data scientist in data science slash machine learning engineer roles for probably a little over a decade now.
45 00:22:13.420 ⇒ 00:22:26.960 Ujval Kamath: Almost… a lot of that has been, like, customer-facing, so I’ve, you know, in what you typically call services, or mixed roles of services and pre-sales as well. So sales and services, generally.
46 00:22:28.620 ⇒ 00:22:39.350 Ujval Kamath: with the exception of, like, I was once a data science manager briefly, but generally speaking, the expectation in all my jobs has been that I sort of work with
47 00:22:39.460 ⇒ 00:22:48.820 Ujval Kamath: you can call them a customer or a stakeholder, whatever you want to call them, and they have some vague kind of data science-y thing they want to do. Oh, we have some data, and we want to do some…
48 00:22:48.820 ⇒ 00:22:49.250 Uttam Kumaran: Yeah.
49 00:22:49.470 ⇒ 00:23:06.469 Ujval Kamath: Yeah, something very vague. And it’s my kind of responsibility to understand what is it really trying to… what are they really trying to do? What are they really asking for? Is it feasible? Is the outcome, like, reasonable? You know, everybody feels like, hey, if I throw my data in this AI bucket, or my ML bucket, or the big data bucket, whatever it was.
50 00:23:06.470 ⇒ 00:23:06.930 Uttam Kumaran: Yeah, yeah.
51 00:23:06.930 ⇒ 00:23:11.719 Ujval Kamath: you know, I’ll generate a trillion dollars. And my job is to sort of say, is this…
52 00:23:11.840 ⇒ 00:23:24.560 Ujval Kamath: you know, is what you’re trying to do reasonable? And then, I sort of build it out as well. So I’ve done that with on-premise stuff, I’ve done that with cloud, you know, using cloud… cloud as well. I think the only part where I…
53 00:23:24.890 ⇒ 00:23:31.630 Ujval Kamath: I haven’t done a lot of is what you’d call typically front-end, so I have done
54 00:23:31.760 ⇒ 00:23:48.039 Ujval Kamath: with BI tools. So Tableau, Power BI. So typically, if someone says, hey, we want to seize the data science part in… if we want to see the prediction somewhere, I would probably do it in a BI tool, and somebody else would do the front end, if they had to want to make a custom dashboard.
55 00:23:48.250 ⇒ 00:23:48.630 Uttam Kumaran: Okay.
56 00:23:48.630 ⇒ 00:23:51.170 Ujval Kamath: That’s changing a little bit with LLMs.
57 00:23:51.590 ⇒ 00:23:58.340 Ujval Kamath: So… I have… I’m doing some work for a friend who’s there trying to bootstrap their own…
58 00:23:58.730 ⇒ 00:24:06.439 Ujval Kamath: a small thing in the manufacturing space, and I can do a front-end, I can do a dashboard in LLM, like, it doesn’t.
59 00:24:06.440 ⇒ 00:24:06.900 Uttam Kumaran: Yeah.
60 00:24:07.260 ⇒ 00:24:10.190 Ujval Kamath: You know, as long as I have the data somewhere.
61 00:24:10.190 ⇒ 00:24:27.429 Uttam Kumaran: I can just go to an LLM and say, this is exactly what I want to see. Well, as data people, we have some basic dashboarding experience, because, you know, and so, like, it’s less about that, it’s actually just, like, drag and drop UI-based flows is so annoying that it’s, like, that’s what’s preventing, commonly.
62 00:24:27.430 ⇒ 00:24:27.790 Ujval Kamath: Yeah.
63 00:24:27.790 ⇒ 00:24:28.450 Uttam Kumaran: Wow.
64 00:24:29.240 ⇒ 00:24:39.860 Ujval Kamath: So, I could… I guess I can call myself more of a full-stack data scientist now, just because I can use an LLM to make, right, to do the actual code for the… for the dashboard, so…
65 00:24:39.860 ⇒ 00:24:40.450 Uttam Kumaran: Yeah.
66 00:24:40.710 ⇒ 00:24:41.330 Ujval Kamath: Yeah.
67 00:24:41.820 ⇒ 00:24:45.870 Ujval Kamath: So, and I mean, where is the line in terms of,
68 00:24:46.020 ⇒ 00:24:50.920 Ujval Kamath: one of the things, like, in all my jobs is I’m… I do like being close to, sort of, the end product.
69 00:24:51.310 ⇒ 00:25:07.180 Ujval Kamath: So, you know, I can… I can understand, like, what is the outcome? Why are we building this? Like, what’s the business value? Which is one of the reasons that I… like I said, I’ve always liked customer work. So when I saw what you were doing, I said, okay, it seems like, to a large degree, they’re, like, a lot closer to their customers and the end product than where…
70 00:25:07.180 ⇒ 00:25:07.860 Uttam Kumaran: Yes.
71 00:25:07.860 ⇒ 00:25:15.930 Ujval Kamath: you know, if you’re in a very large organization, even in a very large services organization, you’re sort of buried somewhere where it’s like, okay, here’s your JIRA ticket.
72 00:25:15.930 ⇒ 00:25:16.770 Uttam Kumaran: Yes, exactly.
73 00:25:16.770 ⇒ 00:25:19.290 Ujval Kamath: Yeah, yeah. So that’s why I…
74 00:25:19.290 ⇒ 00:25:22.719 Uttam Kumaran: to even double down on your point, I, you know, I was always…
75 00:25:22.900 ⇒ 00:25:41.670 Uttam Kumaran: in the data team, and I started as a data engineer. I did a lot of modeling work, I’ve done BI work, I’ve also done a lot of pipelining, work. And so, for me, it’s like… but I always got relegated to, like, the engineering world. But as data people, we tend to know the ecosystem super, super well.
76 00:25:41.670 ⇒ 00:25:42.020 Ujval Kamath: Yeah.
77 00:25:42.020 ⇒ 00:25:49.890 Uttam Kumaran: yet we’re not… don’t have a seat at the table to, like, give our opinion about data. So, when we started this business, we really tried to push for the fact that, like.
78 00:25:49.890 ⇒ 00:26:03.799 Uttam Kumaran: We don’t… we want to go and be a partner to the business, and that’s why we not only do a lot of the data engineering modeling work, we also do strategy and recommendations. We kind of, like, go the full breadth, and so we’re putting together decks, memos.
79 00:26:03.800 ⇒ 00:26:10.410 Uttam Kumaran: and being able to go to board meetings and things like that for our clients. And our advantage is that we are, like, kind of
80 00:26:10.410 ⇒ 00:26:30.100 Uttam Kumaran: pretty vertically integrated in that way. But we also don’t pitch to the CTO’s office, typically. We usually go through the business. And, there’s pros and cons to that, but I always know that, like, we work well with engineers, so I’m never worried about convincing them that we know what we’re doing. It’s more about, like, our recommendations need to go impact
81 00:26:30.130 ⇒ 00:26:34.850 Uttam Kumaran: business changes. Otherwise, it’s a complete waste of money to work with us, you know?
82 00:26:36.020 ⇒ 00:26:38.950 Ujval Kamath: Yeah, that makes… I mean, I definitely understand that, so…
83 00:26:39.880 ⇒ 00:26:51.709 Uttam Kumaran: Yeah, perfect. I mean, tell me about, like, what’s been, like, interesting in the past, like, 5 years in data? Like, what… interesting to you? You know, like, what… I saw a lot of your background in…
84 00:26:51.710 ⇒ 00:26:52.060 Ujval Kamath: Church.
85 00:26:52.060 ⇒ 00:26:54.859 Uttam Kumaran: In data science, I’m sort of curious…
86 00:26:55.180 ⇒ 00:27:05.709 Uttam Kumaran: I’m data science adjacent, I have a lot of data science friends, I’ve seen good and bad data science, but again, I’m not a data scientist myself. I’m sort of curious…
87 00:27:06.280 ⇒ 00:27:20.950 Uttam Kumaran: selfishly about, like, what has changed over the last few years, given LLMs, and given AI, but also just, like, hearing about, like, what you’ve loved about being in this industry as a data scientist.
88 00:27:21.970 ⇒ 00:27:26.409 Ujval Kamath: I mean, so, so, I mean… I like to code.
89 00:27:26.610 ⇒ 00:27:37.850 Ujval Kamath: I mean, I don’t like… I don’t mean, like, an engineer who’s just, like, 8 hours a day, like… that’s their dream, is to just, like, sit at the screen. But I like to build stuff. But I also like the business value part.
90 00:27:37.970 ⇒ 00:27:47.789 Ujval Kamath: So I was, you know, I was never, like, the best, I would say, computer scientist, where you’re like, okay, optimize, make this code, like, 1% faster, you know? I was never really that…
91 00:27:47.880 ⇒ 00:28:06.089 Ujval Kamath: So data science in that sense, it spans those two bridges. Like, a lot of data science roles. You’re not building something just because it’s technologically cool, or interesting. You’re building something where, like, what’s the end result of this? Like, who’s going to actually do something with it? In terms of, what’s interesting.
92 00:28:06.440 ⇒ 00:28:08.030 Ujval Kamath: One is…
93 00:28:08.230 ⇒ 00:28:21.859 Ujval Kamath: I think, at least for me, is how much easier it is for me to do things now. So, I do remember when I started out, you know, every… if you had a lot of data, you were like, oh, Hadoop, and MapReduce, and Spark, and stuff.
94 00:28:22.190 ⇒ 00:28:32.539 Ujval Kamath: I think a lot of people still do that, but the amount of stuff I can do with… just literally on my laptop with some optimized libraries or, like, DuckDB, I’m kind of.
95 00:28:32.540 ⇒ 00:28:42.379 Uttam Kumaran: Are you using DuckTV to basically host local… I think you’re totally right, though, in the way you describe it. People are still doing that, it’s just because the knowledge is not distributed, like.
96 00:28:42.410 ⇒ 00:28:59.459 Uttam Kumaran: folks are just not, like, on Twitter or, like, seeing DuckTV, and it’s, like, I, like, almost, like, I have to explain to you, like, what magic is. Like, you can run basically so much locally, and the availability for, like, just to spin up compute if you need it really easily is, like.
97 00:28:59.460 ⇒ 00:29:00.090 Ujval Kamath: Yeah, it’s just…
98 00:29:00.090 ⇒ 00:29:02.040 Uttam Kumaran: So different. Yeah, so different.
99 00:29:02.610 ⇒ 00:29:10.220 Ujval Kamath: I mean, I’ve been in my previous jobs where people were, like, so concerned with the amount of spend, because, like, oh, we have a Spark cluster and all that stuff.
100 00:29:10.460 ⇒ 00:29:13.499 Ujval Kamath: And sometimes if you tell them, like, listen, I can now do this on, like.
101 00:29:13.500 ⇒ 00:29:15.230 Uttam Kumaran: Only anything.
102 00:29:15.560 ⇒ 00:29:22.050 Ujval Kamath: And it’s… and it’s much faster for me to develop it as well. A lot of people, like, they almost don’t believe it. They’re almost like, oh, no, no, we haven’.
103 00:29:22.050 ⇒ 00:29:36.789 Uttam Kumaran: Well, it’s like, how do you talk about, yeah, how do you explain magic? Like, you know? So, it’s the same thing when we were pitching AI services, like, two years ago, people just didn’t get it. And, like, I’ve been using… we started using GPT 3.5, and, like.
104 00:29:36.830 ⇒ 00:29:43.300 Uttam Kumaran: so AI-embedded, it was only until this, like, 6 months ago were we able to actually, like.
105 00:29:43.390 ⇒ 00:30:02.130 Uttam Kumaran: people are like, okay, I’ve tried something, and I’ve seen where, like, ChatGPT hits a wall, and so we need to talk about, like, integrations, context engineering. Okay, perfect. But until then, it wasn’t really possible, so… yeah, that’s, yeah, that’s… it makes sense.
106 00:30:02.130 ⇒ 00:30:02.700 Ujval Kamath: Yeah.
107 00:30:03.100 ⇒ 00:30:05.140 Ujval Kamath: And, and of course, like,
108 00:30:05.250 ⇒ 00:30:19.619 Ujval Kamath: Llms are… I mean, everything around LLMs are super interesting, of course. I… I think even more interesting, like, whenever I work in an LLM, anything with LLM, I’m always surprised where it breaks in, like, the weirdest ways. You know, it just…
109 00:30:19.830 ⇒ 00:30:39.770 Ujval Kamath: I think that’s interesting, because it’s not something that a lot of people, I think, think about, is that if you’re technical, you understand that LLMs, or even agents, they do weird, weird, weird, weird things sometimes that you don’t expect. It’s not something… it’s just interesting to me that it’s, like, failed sometimes in the most obvious ways, and it’s just, like, I’m just curious where things are gonna go.
110 00:30:39.880 ⇒ 00:30:41.719 Ujval Kamath: And I think everyone is.
111 00:30:41.850 ⇒ 00:30:47.790 Ujval Kamath: I think in terms of deep learning and AI, what’s really interesting to me is, you know, for machine learning.
112 00:30:48.110 ⇒ 00:30:51.580 Ujval Kamath: There’s… you know, I always used to think of this, like, tabular data.
113 00:30:52.020 ⇒ 00:31:04.159 Ujval Kamath: you know, it’s like, oh, everything is in, like, this form of X and Y. Now, it’s like, you can have any sort of arbitrary input into a neural network, and I still, like, struggle so much to, like, believe it.
114 00:31:04.640 ⇒ 00:31:05.720 Ujval Kamath: That…
115 00:31:05.740 ⇒ 00:31:18.289 Ujval Kamath: you know, when you see these models where, like, you look, and they’re like, oh, we fed in text, and we fed in images, and you can feed any random thing into, like, a neural network, and it’ll feed something… you’ll get something out of it. I still struggle so much with, like, that mental transition that
116 00:31:18.290 ⇒ 00:31:26.809 Ujval Kamath: You can literally stuff anything into a neural network, as long as you, like, define the inputs properly. And it’s just so different from what, like, historically, people thought about data.
117 00:31:26.810 ⇒ 00:31:27.660 Uttam Kumaran: Yeah.
118 00:31:28.110 ⇒ 00:31:30.300 Uttam Kumaran: I mean, even, like, even the last 10 years.
119 00:31:30.450 ⇒ 00:31:31.730 Uttam Kumaran: Yeah.
120 00:31:32.430 ⇒ 00:31:33.240 Uttam Kumaran: Yeah.
121 00:31:34.710 ⇒ 00:31:36.850 Ujval Kamath: I mean, I… I…
122 00:31:37.520 ⇒ 00:31:46.490 Ujval Kamath: you can feed, like… I have tried this, like, you feed, like, a book, like, a children’s book into, like, an LLM, and be like, describe every page to me, and somehow it, like.
123 00:31:46.680 ⇒ 00:31:51.219 Ujval Kamath: like, extract everything. I just don’t even understand how that works. It’s just nuts.
124 00:31:51.410 ⇒ 00:31:55.019 Uttam Kumaran: Yeah, and again, it’s changing every 3 to 6 months.
125 00:31:55.180 ⇒ 00:32:00.710 Uttam Kumaran: It’s, like, almost doubling… at minimum doubling in every core metric.
126 00:32:01.010 ⇒ 00:32:07.299 Uttam Kumaran: And it’s getting cheaper. It’s, like, not… it’s, like, really hard… we don’t work this fast, like, we’re, like…
127 00:32:07.630 ⇒ 00:32:08.390 Ujval Kamath: No, no, I mean…
128 00:32:08.390 ⇒ 00:32:14.830 Uttam Kumaran: Like, for example, I got an email that they’re sunsetting, like, an Opus 4, and, like.
129 00:32:14.830 ⇒ 00:32:15.300 Ujval Kamath: I don’t know.
130 00:32:15.300 ⇒ 00:32:17.899 Uttam Kumaran: I do, it just came out, like, 6 months ago.
131 00:32:17.900 ⇒ 00:32:19.600 Ujval Kamath: Yeah, exactly.
132 00:32:20.340 ⇒ 00:32:24.179 Ujval Kamath: It’s a… it’s… it’s just… it’s so fast now that it’s hard for…
133 00:32:24.680 ⇒ 00:32:42.439 Ujval Kamath: It’s just, yeah, it’s hard to understand, and you know, little, little things, like, you know, I’ve certainly seen this one. I use an LLM, I’m like, okay, LLMs are not there yet to do this particular problem, and then they’re like, oh no, it’s just that you’re using, like, this two-day-old version of something on Amazon, whereas if you use this other model, it can do that.
134 00:32:42.440 ⇒ 00:32:45.480 Uttam Kumaran: You’re like, oh, you guys just didn’t use the right one, yeah.
135 00:32:45.480 ⇒ 00:32:47.940 Ujval Kamath: Yeah, it’s hard to… I just don’t know how people keep up with.
136 00:32:47.940 ⇒ 00:32:52.440 Uttam Kumaran: But most of it, most of what we’re finding, it’s actually a data engineering
137 00:32:52.560 ⇒ 00:33:06.220 Uttam Kumaran: and an integrations problem, more than picking the right LLM. And, like, any phone that’s smart can do the prompt engineering, but getting, like, doing the ETL to get structured data into a point at which you can
138 00:33:06.290 ⇒ 00:33:21.009 Uttam Kumaran: point an LLM to it, and taking the result, like, whatever the result, whether it’s structured outputs or unstructured, and getting it to, like, Slack or to another application. That’s, like, what… most of, actually, what we’re doing for people.
139 00:33:21.010 ⇒ 00:33:21.790 Ujval Kamath: Makes sense.
140 00:33:21.790 ⇒ 00:33:23.620 Uttam Kumaran: But, again, the magic is, like.
141 00:33:23.760 ⇒ 00:33:29.340 Uttam Kumaran: we sort of just, like, pick the best magic, like, but again, it’s not actually the limiting factor. It’s like…
142 00:33:29.450 ⇒ 00:33:33.049 Uttam Kumaran: oh, like, these don’t have APIs, or, like, you guys don’t know Warehouse, like.
143 00:33:33.290 ⇒ 00:33:39.960 Uttam Kumaran: We’re not going to be able to solve this until you do a lot of that work, which is pulling a lot of, like, our business forward, you know, frankly.
144 00:33:41.720 ⇒ 00:33:49.959 Ujval Kamath: Yeah, which makes sense, and I mean, I think even large companies, they always think their data… the shape of their data systems and their data is always much worse than they think it is, so…
145 00:33:50.250 ⇒ 00:33:50.810 Uttam Kumaran: Yeah.
146 00:33:51.800 ⇒ 00:33:54.850 Uttam Kumaran: And so, tell me, I guess, like, I’m interested in, sort of, like.
147 00:33:54.950 ⇒ 00:34:09.959 Uttam Kumaran: you know, maybe I can tell you a little bit about… I’m interested in sort of some of the work that you did, and sort of, you know, you mentioned, even through your, you know, I know you kind of have your own firm now, but I’m sort of interested in the kind of work that you’re interested in, even, like, how…
148 00:34:09.960 ⇒ 00:34:17.979 Uttam Kumaran: business stakeholders perceive it, because for me, I’m… now I’m getting… we’re an expert in, like, why does a data warehouse matter to moving
149 00:34:18.070 ⇒ 00:34:27.230 Uttam Kumaran: your XKPI, right? And so, I’m interested in… this would be our first time, sort of, you know, working with somebody in the data science. I think a lot of our
150 00:34:27.440 ⇒ 00:34:29.920 Uttam Kumaran: Internal, like.
151 00:34:30.190 ⇒ 00:34:44.829 Uttam Kumaran: our internal, you know, we say capabilities really end at the, like, more advanced data engineering piece, but I’ve always told our team that as we get to bigger and bigger clients, right, they actually tend to have some of this figured out, and they actually do need
152 00:34:44.909 ⇒ 00:34:54.099 Uttam Kumaran: data science, and they do need some more thinking there. And again, for me, as I told you, like, our clients like us because we’re sort of becoming a one-stop shop.
153 00:34:54.120 ⇒ 00:35:01.109 Uttam Kumaran: in everything in data, and it’s not that we do things where we’re not gonna do it to the A-plus degree.
154 00:35:01.110 ⇒ 00:35:16.020 Uttam Kumaran: So that’s the thing, like, if a client today comes to me and asks for data science support, I said, I’m not… I can’t do that, right? Because we don’t have those people, and we’re not… one thing that’s helped us in this business, we’ve never lowered the bar on what we give to clients. In fact.
155 00:35:16.020 ⇒ 00:35:28.860 Uttam Kumaran: Where we’ve always struggled, because most of my team are product engineers, is just the communication piece. So we learn how to, like, communicate to the client better. Really, that’s, like, the consulting, but most of our team are engineers, so we’ve never compromised the quality of our work.
156 00:35:28.960 ⇒ 00:35:41.320 Uttam Kumaran: And so really, like, I was… I’m happy to talk to you, because I was interested in, sort of, like, how you’ve seen throughout your career communicating the ROI in data science, but even, like, where ML…
157 00:35:41.610 ⇒ 00:35:46.040 Uttam Kumaran: You know, modeling, or deep learning modeling, or advanced statistical modeling, like.
158 00:35:46.270 ⇒ 00:35:50.229 Uttam Kumaran: what is the story to the executive? You know, I’m kind of curious, like, what you think about that.
159 00:35:51.280 ⇒ 00:36:03.010 Ujval Kamath: Sure, so, I mean, first of all, I’ve never… I mean, I long time… a long time ago, I stopped talking about deep learning. I mean, so, you know, I… we just bunch it into, like, whatever. I think…
160 00:36:03.180 ⇒ 00:36:10.419 Ujval Kamath: You just bunch into whatever the latest term is. So, you know, predictive analytics, proactive analytics.
161 00:36:10.420 ⇒ 00:36:10.780 Uttam Kumaran: Sure.
162 00:36:10.780 ⇒ 00:36:11.669 Ujval Kamath: I don’t think.
163 00:36:11.670 ⇒ 00:36:20.020 Uttam Kumaran: Yeah, I know, I even use a bunch of terms, so that’s even more important to me, is to hear what is… what’s working for your messaging, too.
164 00:36:20.020 ⇒ 00:36:27.129 Ujval Kamath: Yeah, I don’t think I’ve ever heard, like, anybody, like a business stakeholder, an executive, really, really care about, like.
165 00:36:27.570 ⇒ 00:36:32.700 Ujval Kamath: The underlying technology, or the… you know, we use data science, we use predictive analytics.
166 00:36:32.700 ⇒ 00:36:36.600 Uttam Kumaran: They may care with the logo, they may be like, oh, are we using Databricks or Snowflake? Yeah, exactly.
167 00:36:36.600 ⇒ 00:36:37.920 Ujval Kamath: Beautiful.
168 00:36:37.920 ⇒ 00:36:40.470 Uttam Kumaran: Don’t worry too much about that.
169 00:36:40.470 ⇒ 00:36:55.919 Ujval Kamath: Yeah, occasionally I’ll find, like, a stakeholder who’s, like, taking a data science course, and then they’ll ask me, like, oh, what model are you using? But to a large degree, I think what I… I think what I have done a lot in my job is try to sort of map
170 00:36:56.150 ⇒ 00:37:13.249 Ujval Kamath: the business value part of it. I’ll give you an example with, like, Siemens… I was at Siemens Healthineers, they make medical devices, and I’ll talk to you a little bit about, like, the outcome over there. So, Siemens Healthineers, they make medical devices, but they also service them.
171 00:37:13.510 ⇒ 00:37:19.810 Ujval Kamath: So you can imagine, like, you want to go to a hospital, and you want to, you know, you…
172 00:37:19.970 ⇒ 00:37:28.580 Ujval Kamath: you’re getting a blood test, right? Your blood test goes to a facility which actually, like, puts that blood in a machine that does all the tests.
173 00:37:28.640 ⇒ 00:37:29.340 Uttam Kumaran: Yeah.
174 00:37:29.560 ⇒ 00:37:46.899 Ujval Kamath: The machine needs to work, right? You don’t want to be in a situation where you’re a facility, and you’re, like, you know, your revenue stream is, we do blood tests, and we have… that’s how we make money, is, like, by doing these blood tests and giving people results. You wake up on Monday morning, and your machines are all broken.
175 00:37:47.050 ⇒ 00:37:57.359 Ujval Kamath: Right? That you lost your revenue stream because you can’t test any blood. So what Siemens also did is they basically, I always say, they sold an extended warranty on the machine they.
176 00:37:57.360 ⇒ 00:37:57.770 Uttam Kumaran: Okay.
177 00:37:57.810 ⇒ 00:38:02.530 Ujval Kamath: So, if the machine broke, they would go in and they would fix it.
178 00:38:02.880 ⇒ 00:38:06.590 Ujval Kamath: And then, you know, what they do is, instead of saying.
179 00:38:07.040 ⇒ 00:38:14.500 Ujval Kamath: okay, every time you fix it, you pay… I’m gonna give you a bill. They say, you pay us, like, an annual fee, and we will go in and we will do whatever repairs we did.
180 00:38:14.630 ⇒ 00:38:19.000 Ujval Kamath: Siemens, of course, had a service cost for doing those repairs.
181 00:38:19.150 ⇒ 00:38:37.240 Ujval Kamath: It was really, really high. So you can imagine you have all these people, they’re going on site, they’re repairing machines, they’re changing, like, really expensive parts sometimes. Sometimes the repairs can take, like, 8 hours or 2 days. There’s a very, very high, like, just time and materials cost. It’s like a human cost to doing that.
182 00:38:37.260 ⇒ 00:38:42.590 Ujval Kamath: And what they said was something like, hey, you know, we have an issue.
183 00:38:43.000 ⇒ 00:38:46.840 Ujval Kamath: If this breaks at a customer site, it costs us, like.
184 00:38:47.020 ⇒ 00:38:52.509 Ujval Kamath: $10,000 to send, like, people out there to fix it. What if we could do something where…
185 00:38:52.890 ⇒ 00:39:02.210 Ujval Kamath: we could predict that issue is gonna happen. We can do it in a way where we can predict it, like, a month in advance.
186 00:39:02.850 ⇒ 00:39:09.650 Ujval Kamath: We can then, call them and say, hey, we’re gonna send one person out on the weekend,
187 00:39:10.220 ⇒ 00:39:12.060 Ujval Kamath: What is the cost savings to us?
188 00:39:12.740 ⇒ 00:39:22.489 Ujval Kamath: So, it’s like, you just think about it, like, I’m giving an analog, which is, like, call an HVAC person when your AC bursts and, like, breaks in, like.
189 00:39:23.070 ⇒ 00:39:26.510 Ujval Kamath: the worst day in Austin. Like, it’s like 120 degrees.
190 00:39:26.510 ⇒ 00:39:30.789 Uttam Kumaran: I’d be surprised, we have… one of our clients is a home services company, and yes.
191 00:39:30.790 ⇒ 00:39:31.200 Ujval Kamath: better.
192 00:39:31.200 ⇒ 00:39:36.729 Uttam Kumaran: They… they… their… their challenge is that all their customers, they don’t call proactively.
193 00:39:36.730 ⇒ 00:39:37.140 Ujval Kamath: way.
194 00:39:37.140 ⇒ 00:39:42.479 Uttam Kumaran: Right? But they do have maintenance plan to prevent this. It’s a very… I totally get where you’re coming from.
195 00:39:42.710 ⇒ 00:39:53.950 Ujval Kamath: Yeah, and you know, if you… like, I’ve been in the situation where my, like, my heater stopped working on, like, the worst day in winter, and somebody was like, I absolutely could send somebody out today on a Saturday, but it’s gonna cost you, like, double.
196 00:39:54.260 ⇒ 00:39:56.169 Ujval Kamath: Versus if I’d sent them, like.
197 00:39:56.170 ⇒ 00:40:03.129 Uttam Kumaran: Well, so, for example, there’s a storm that came through Austin, and guess what? It’s like, for the tree trimmers, it’s like the best, because it’s Christmas, right?
198 00:40:03.130 ⇒ 00:40:03.980 Ujval Kamath: Exactly.
199 00:40:03.980 ⇒ 00:40:04.560 Uttam Kumaran: Yeah.
200 00:40:04.560 ⇒ 00:40:13.820 Ujval Kamath: So, so then, you know, what I did, like, not just me, but, like, what we did as a data science team is, like, we work with this sort of serv… what they call the service people.
201 00:40:13.910 ⇒ 00:40:28.649 Ujval Kamath: And you map, and you say, okay, listen, if I get you one positive prediction, what’s that worth? And the business stakeholders kind of have an understanding of this. It’s like, oh, in theory, like, we know exactly how much this costs us after the… to send in somebody to repair it when the machine breaks.
202 00:40:28.860 ⇒ 00:40:35.130 Ujval Kamath: We have, like, sort of an estimate of what it would cost us if we were able to, like, do this in a much more optimized way.
203 00:40:35.370 ⇒ 00:40:41.260 Ujval Kamath: And we say, okay, so then I would look through the historical data, and I would say, I’ve seen this happen
204 00:40:42.250 ⇒ 00:40:45.430 Ujval Kamath: I don’t know, like, 500 times in the past year.
205 00:40:45.610 ⇒ 00:40:50.589 Ujval Kamath: If we could predict 80 of those, that’s the cost savings.
206 00:40:50.590 ⇒ 00:40:52.950 Uttam Kumaran: And, well, how did you arrive… how did you arrive at 80?
207 00:40:53.630 ⇒ 00:40:54.879 Ujval Kamath: I mean…
208 00:40:54.880 ⇒ 00:40:58.169 Uttam Kumaran: conservative guess? That’s just being conservative guess.
209 00:40:58.170 ⇒ 00:41:16.220 Ujval Kamath: conservative guess, and you try to come up with… I always say, like, depending on the size of the company, you come up with a number which, like, entices the executives, right? If you’re a billion-dollar company, and you say, I’m going to save you $250,000 a year, they’re gonna be like, we don’t care, we’re not going to invest in that. Your one data scientist cost me more than that.
210 00:41:17.030 ⇒ 00:41:23.050 Ujval Kamath: You say, okay, we’re gonna save you 30 million a year, whatever.
211 00:41:23.220 ⇒ 00:41:41.219 Ujval Kamath: But that’s exactly how you talk… you talk to these executives, is you… you say, if we have this sort of predictive program in thing, and we… we sort of target your top 20 issues, we sort of theoretically… like, we can save this much money.
212 00:41:41.380 ⇒ 00:41:46.600 Ujval Kamath: Then do a pilot to be like, let’s… okay, let’s say we said 80%. Is that realistic?
213 00:41:46.600 ⇒ 00:41:46.970 Uttam Kumaran: Yeah.
214 00:41:46.970 ⇒ 00:41:51.609 Ujval Kamath: And I’ve absolutely been in a condition… I’ve actually been in a situation where, like, look at 20 issues.
215 00:41:51.720 ⇒ 00:41:59.040 Ujval Kamath: 3 of them, we absolutely can hit 80%. Five of them, there’s no way, the data’s so bad. Then you revise your estimates, you know?
216 00:41:59.450 ⇒ 00:42:02.269 Ujval Kamath: Depending on the organization.
217 00:42:02.780 ⇒ 00:42:09.060 Ujval Kamath: like, I’ll be honest, like, when I was at Siemens, I think they were pretty good about being like, what is, like, the realistic number?
218 00:42:09.240 ⇒ 00:42:14.770 Ujval Kamath: When I was a GM, it was all, like, pie-in-the-sky stuff. I mean…
219 00:42:15.000 ⇒ 00:42:20.570 Ujval Kamath: even if I, like, communicated the numbers realistically, like, up to somebody, they would then, like, Finds the number
220 00:42:20.970 ⇒ 00:42:22.670 Ujval Kamath: Yeah, bullshit up today.
221 00:42:22.670 ⇒ 00:42:25.560 Uttam Kumaran: So that’s a… the other… to, again, tell you a little bit of, like.
222 00:42:25.800 ⇒ 00:42:33.010 Uttam Kumaran: it’s so nice, because as data people, we’re all… I know what it’s like to be like, make the graph go up.
223 00:42:33.750 ⇒ 00:42:34.510 Ujval Kamath: Yeah.
224 00:42:34.510 ⇒ 00:42:48.640 Uttam Kumaran: And I really did not like that, and I don’t like working in environments like that. And so, lovely thing about our business is that we don’t need to play internal company politics, because we are an outside stakeholder. We are, of course.
225 00:42:48.830 ⇒ 00:42:59.620 Uttam Kumaran: trying to make our stakeholder internally famous, but at the same time, we’re here to do what we know is best and provide a recommendation. So we’re not… we don’t have to worry, like.
226 00:42:59.620 ⇒ 00:43:16.300 Uttam Kumaran: yes, they hired us, they can also fire us, but it’s not gonna be because we didn’t play the politics, you know, or we’re not in for a promotion, like, you know, so that’s what I like about this, is that… is that we… we get to skip a lot of that, which I knew. It’s like, that’s all a lot of my career was, like, figuring out
227 00:43:16.350 ⇒ 00:43:24.460 Uttam Kumaran: how to convince people, even when I knew the answer, you know, and so we don’t have to do a lot of that. The other thing that’s fun is that
228 00:43:24.990 ⇒ 00:43:43.059 Uttam Kumaran: we’re working, as I mentioned, in home services, in B2C SaaS, in legal, in, like, vitamins and nutrition, and the data problems, yes, the end customer and the… their end customer are different, but, like, we’re learning a lot.
229 00:43:43.170 ⇒ 00:43:57.210 Uttam Kumaran: And we’re learning a lot, and the next customer is benefiting from the fact that we’ve seen so much. And not only what we’ve seen at Brainforge, every team member has done, you know, 3 to 5 gigs before this.
230 00:43:57.210 ⇒ 00:44:06.719 Uttam Kumaran: and is, like, brings a wealth. And so, that’s just, like, compounding. And so, internally at our company, we operate more like a product company, in that there’s a lot of cross…
231 00:44:06.990 ⇒ 00:44:14.600 Uttam Kumaran: platform, like, all the teams are sort of talking to each other. It’s… I’ve worked in consulting where it’s, like.
232 00:44:14.610 ⇒ 00:44:33.219 Uttam Kumaran: you just… you just know who’s on your project. Like, you don’t even know that… someone just… you just do a timesheet at some portal, and, like, this is not that. This is, like, a real team, where we are… our product is this service, and we are deploying that to multiple customers, but internally, we run, like, a product
233 00:44:33.460 ⇒ 00:44:42.680 Uttam Kumaran: like a product where it’s a lot of shared learnings. It’s serv… like, we have different services that we offer, and those service lines and all the leaders within them meet.
234 00:44:42.880 ⇒ 00:44:58.890 Uttam Kumaran: And then we have, like, entire team, entire data team, entire AI team, so it’s so much collaboration, and that we’re able to build playbooks and build, like, ways of doing things that get shared among all the people at Brainforge, and then immediately go to help the next
235 00:44:58.890 ⇒ 00:45:07.479 Uttam Kumaran: and existing, you know, customers, which is, again, this is, I think, what is fun about this company compared to any consulting or companies that I’ve worked at.
236 00:45:07.550 ⇒ 00:45:25.860 Uttam Kumaran: in general, is that, like, we have this really boring service industry component, right, where we’re doing IT consulting. But, like, for me, what’s fun is that we internally are thinking, okay, we’ve just done things for one client here. We’ve done the exact same thing again. Okay, we need to start to, like.
237 00:45:25.970 ⇒ 00:45:45.509 Uttam Kumaran: codify that into some playbook. And that allows us to not only, like, yes, our margins are better, but also we’re not doing, like, busy work. Like, all of our team are doing, like, really the work where I tell our client, your job, you should throw us into the fire. Like, having us do rudimentary data cleaning is not a good use of your funds.
238 00:45:45.510 ⇒ 00:45:59.359 Uttam Kumaran: happy to do it, but you want to use us like Navy SEALs, you know? Throw us into, like, really the tough, time-pressing problems. And people don’t call us because things are going well, they call us because they don’t have visibility, or, like.
239 00:45:59.660 ⇒ 00:46:18.110 Uttam Kumaran: something happened where they, like, need a fix, you know? But immediately we come in, we deliver that, and then they’re like, what else can you do for us? You know? So that’s been… that’s what’s been really, really fun. I guess I might… maybe one more curious question. I know we’re actually way over time, by the way, so I don’t know if you’re okay, but .
240 00:46:18.110 ⇒ 00:46:18.840 Ujval Kamath: Oh, I’m okay.
241 00:46:18.840 ⇒ 00:46:29.599 Uttam Kumaran: Okay, I was gonna ask about, like, how you think about the types of companies that can get value from, like, data science.
242 00:46:29.760 ⇒ 00:46:36.609 Uttam Kumaran: And the types of companies who may ask for data science and can’t… and, like, you’re, like, you guys are, like.
243 00:46:36.750 ⇒ 00:46:43.599 Uttam Kumaran: either too small, or it’s too crazy, or you don’t have the right thinking. To give you an example, I’ve always been of the belief that, like.
244 00:46:43.990 ⇒ 00:46:48.690 Uttam Kumaran: for most companies, SQL can get you, like, 95% of the way back.
245 00:46:48.690 ⇒ 00:46:49.080 Ujval Kamath: I agree.
246 00:46:49.080 ⇒ 00:46:59.880 Uttam Kumaran: I would agree. I have a lot of data science friends, I love them, but sometimes I’ve seen data science projects go for 2 years, and I’m like, wow, I could have just
247 00:46:59.980 ⇒ 00:47:18.920 Uttam Kumaran: dodge a cohort analysis for you, like, in, like, an afternoon. So I am interested to hear about, like, the… what is the ideal customer for, you know, some of the wins that you described? It… it doesn’t need to be revenue, it doesn’t need… like, but I’m kind of interested in, like, how you think about that question.
248 00:47:19.470 ⇒ 00:47:30.820 Ujval Kamath: I mean, so, I maybe split that into two things. So, one is, I do agree with you, like, I can’t tell you how many data science… you know, data science projects I worked on where, like.
249 00:47:31.250 ⇒ 00:47:32.980 Ujval Kamath: I came up with some rules.
250 00:47:33.110 ⇒ 00:47:43.299 Ujval Kamath: And that covered, like, 90% of the cases. And the person was like, oh my god, this is so amazing, you’re able to, like, you know, raise all these things, and I’m like, oh yeah, you know?
251 00:47:43.550 ⇒ 00:47:51.760 Ujval Kamath: statistics. I mean, you don’t see… they don’t care, like, how it’s done, but… I mean… So…
252 00:47:52.160 ⇒ 00:47:54.919 Ujval Kamath: I think there’s two things is,
253 00:47:55.790 ⇒ 00:48:07.590 Ujval Kamath: I actually haven’t seen a lot of small companies… like, unless you are talking about product analytics or something like that, like a software company, I haven’t seen enough small companies where
254 00:48:07.900 ⇒ 00:48:24.890 Ujval Kamath: I think they really have the resources to take what I would call an ML-based approach. So, I’ll give you an example, like, I have a friend who’s… they’re trying to, like, spin something up in, like, the small manufacturer space. So, these are companies that… literally, like, 100 people, they have, like, 100 customers.
255 00:48:26.330 ⇒ 00:48:35.349 Ujval Kamath: Their issue is not that they don’t have, like, they don’t have data. They don’t have anyone to do anything with the result from a data science project.
256 00:48:35.540 ⇒ 00:48:39.480 Uttam Kumaran: Because they can’t… they can’t… they can’t afford a data team, basically. Or, like, it’s not…
257 00:48:39.480 ⇒ 00:48:51.850 Ujval Kamath: It’s not that they… it’s not that they can’t afford a data team. Let’s say they even said, okay, we’ll hire Ujwal, we’re gonna hire you as a contract position. Make something on AWS that sends us some emails, or, like, gives us a dashboard.
258 00:48:52.210 ⇒ 00:49:01.499 Ujval Kamath: I’m just… then it’s like, okay, fine, I’m gonna do that for you. Do you have somebody who’s gonna look at that dashboard every morning and actually do something? And the answer is no, we don’t. We’re like…
259 00:49:01.680 ⇒ 00:49:05.649 Ujval Kamath: We’re, like, dying here just to, like… Keep the lights on.
260 00:49:05.650 ⇒ 00:49:06.000 Uttam Kumaran: Yeah.
261 00:49:06.000 ⇒ 00:49:13.860 Ujval Kamath: We’re not… we can’t invest to do it. We can’t, like, do anything with the results, even if you built, like, a really awesome model that…
262 00:49:13.860 ⇒ 00:49:28.720 Uttam Kumaran: For us, we’re, like, we have some clients like that too, so we spoon, we literally spoon feed, go do this. Like, we spoon-feed the recommendations, but we have, we have clients where, like, one team will be like that, one team will be like.
263 00:49:28.720 ⇒ 00:49:40.850 Uttam Kumaran: I-4 analysts, like, just produce the clean tables, and we’ll go from there, right? But you’re totally right. It’s like, if there’s nobody… and it’s also, it’s a manufacturing company, there’s… you’re not gonna get great data people that want to work at any.
264 00:49:40.850 ⇒ 00:49:42.360 Ujval Kamath: No, no, I mean…
265 00:49:42.360 ⇒ 00:49:52.520 Uttam Kumaran: So, you kind of have to go external, and… but again, for us, like, I won’t take that deal if we’re not gonna succeed. Like, I’m not gonna have our team work
266 00:49:52.600 ⇒ 00:50:02.980 Uttam Kumaran: And then nobody buys into this. And we’ve had this happen, which is… it’s not like I came up with this. We’ve gone through this, like, before, where we did the great job, but nobody’s there to…
267 00:50:03.100 ⇒ 00:50:07.809 Uttam Kumaran: They can never meet us, even halfway or close, and so it’s, like, completely useless.
268 00:50:08.750 ⇒ 00:50:27.850 Ujval Kamath: Yeah, and that’s… that’s definitely, like, an area where, like, if you… you know, I always say, like, BI, you know, BI orgs at large companies, or, like, a BI team, they build, like, 500 dashboards, and, like, 10 of them are used, you know, but what’s the value? The other thing I would say is that, at least my experience with, like, medium and large-sized companies is, like.
269 00:50:30.400 ⇒ 00:50:36.319 Ujval Kamath: when I was in services, I could very quickly say if, like, a project was gonna be worthless.
270 00:50:36.630 ⇒ 00:50:47.980 Ujval Kamath: is if… if they had an actual problem, like, a concrete problem that they could map, like, a dollar value to. I always use examples, like, whenever I worked on marketing projects.
271 00:50:48.490 ⇒ 00:51:04.049 Ujval Kamath: or, like, these kind of social media marketing projects, it was always like, oh, we want to be on Facebook. Like, oh, you know, oh, we want to advertise something on Facebook, or we want to do something on TikTok. I was like, okay, like, what’s the, like, the end result here? You know, what’s, like, what’s…
272 00:51:04.540 ⇒ 00:51:18.580 Ujval Kamath: like, you want to increase your marketing spend on the social media to do what? You know, like, what is the sort of ROI? And the answer is, oh, we don’t know, we just think it would be awesome to be on TikTok or something. We think… we want to convince our management that they should make, like, these 30-second shorts.
273 00:51:19.230 ⇒ 00:51:29.079 Ujval Kamath: This is where, like, I feel like large companies, if there isn’t, like, a very, very, very, very concrete way that you can map, like, sort of the data science outcome to
274 00:51:29.310 ⇒ 00:51:37.460 Ujval Kamath: like, a business outcome. It’s not… I mean, you can do it, you can make a lot of money doing that, like, doing useless analyses, but,
275 00:51:37.650 ⇒ 00:51:39.479 Ujval Kamath: I, I, I, I just don’t think…
276 00:51:39.850 ⇒ 00:51:42.939 Ujval Kamath: It’s something that really works well in the long term. Yeah.
277 00:51:43.100 ⇒ 00:51:48.070 Ujval Kamath: I think smaller companies, it’s a little easier because… album.
278 00:51:48.610 ⇒ 00:51:56.489 Ujval Kamath: the person you’re talking to, like, let’s say they have sort of a bit of a pie-in-the-sky idea, they’re also the person who has to live with the business outcome.
279 00:51:56.950 ⇒ 00:52:00.040 Ujval Kamath: So if they’re like, hey, you know, we want to do some sort of
280 00:52:00.210 ⇒ 00:52:11.839 Ujval Kamath: market… mixed market modeling to understand if we should do it on Facebook. They understand because they’re the ones who have to ultimately pay for, like, the failure of a project. Large company is, like, somebody else’s problem, right, if a project fails?
281 00:52:12.150 ⇒ 00:52:18.229 Ujval Kamath: So I think there is a gray area in there. But, that’s kind of my take on it, like, how to evaluate it.
282 00:52:18.950 ⇒ 00:52:19.520 Uttam Kumaran: Yeah.
283 00:52:20.160 ⇒ 00:52:23.920 Ujval Kamath: If a pro… if something I’m gonna do is actually gonna make a difference to somebody.
284 00:52:23.960 ⇒ 00:52:24.820 Uttam Kumaran: Yeah.
285 00:52:25.660 ⇒ 00:52:29.059 Ujval Kamath: But like I said, data… a lot of data scientists can make a lot of money just.
286 00:52:29.390 ⇒ 00:52:44.189 Uttam Kumaran: Yeah, I know, I know. But it’s not… it’s not… I’m not interested in that type of money. I’m not interested in doing that. Like, I don’t know. I can’t. I never was good at doing BS. It’s sort of why I can’t, like, I kind of left that world, because it’s too much of both, and like…
287 00:52:44.210 ⇒ 00:52:51.140 Uttam Kumaran: there was some great projects I worked on where it was, like, awesome, and then… but in order to get those to happen, I had to do a bunch of crap that didn’t go… it’s like…
288 00:52:51.140 ⇒ 00:52:51.680 Ujval Kamath: Yeah.
289 00:52:51.680 ⇒ 00:52:58.240 Uttam Kumaran: I don’t like that. I… so we don’t play… we just don’t play those games. We don’t take that type of work for clients, typically.
290 00:52:58.890 ⇒ 00:53:11.990 Uttam Kumaran: I’m sort of interested in, like, what you’re, like, what you’re up to now, and, like, what, like, I assume you’re still kind of consulting solo? Like, what is the situation? What are you looking for? Yeah, I’m just curious.
291 00:53:12.400 ⇒ 00:53:17.740 Ujval Kamath: Sure, so, I mean, my… the thing I have on my… my job… my thing, it’s really more of a research project.
292 00:53:17.740 ⇒ 00:53:18.250 Uttam Kumaran: Okay.
293 00:53:18.250 ⇒ 00:53:21.759 Ujval Kamath: it’s not like a paid consulting, but I do try to talk to
294 00:53:21.930 ⇒ 00:53:29.650 Ujval Kamath: people in, sort of, manufacturing and oil and gas and stuff, to just understand, like, business outcomes more so than anything. And then I, like.
295 00:53:30.010 ⇒ 00:53:33.890 Ujval Kamath: I have a blog where I… it’s… there’s still a long-form blog where I kind of map stuff to…
296 00:53:34.280 ⇒ 00:53:38.040 Uttam Kumaran: I’m not encouraging you to read it, but I’m just saying that… I’m gonna read it!
297 00:53:38.040 ⇒ 00:53:43.029 Ujval Kamath: What it means in terms of… what it means in terms of, you know, where some of that stuff is coming from.
298 00:53:43.030 ⇒ 00:53:43.740 Uttam Kumaran: Yeah.
299 00:53:43.960 ⇒ 00:53:51.959 Ujval Kamath: But, I mean, I… I have really sort of been upskilling more on just small, small things like AWS and
300 00:53:52.100 ⇒ 00:53:58.100 Ujval Kamath: bedrock and things like that. That’s kind of what I’ve been doing with my time, because,
301 00:53:59.320 ⇒ 00:54:14.260 Ujval Kamath: whenever I’ve been at a large company… most of my jobs, I’ve been very constrained in the tools I use. I don’t mean that as a bad way, but what tends to happen is, like, oh, it’s somebody else’s job to do this part of the cloud stuff. You… like, I used to use Databricks a lot, but…
302 00:54:14.370 ⇒ 00:54:18.810 Ujval Kamath: I… we had other stuff where I was like, no, no, you don’t touch that, because it’s not your…
303 00:54:18.810 ⇒ 00:54:19.400 Uttam Kumaran: Yeah.
304 00:54:19.680 ⇒ 00:54:25.830 Ujval Kamath: So I just have been messing around with AWS to try to see, like, okay, if I had to build, like, my own product.
305 00:54:25.830 ⇒ 00:54:26.290 Uttam Kumaran: Fantastic.
306 00:54:26.290 ⇒ 00:54:29.789 Ujval Kamath: If I had to build, like, an end-to-end, like, a real end-to-end for a customer.
307 00:54:30.070 ⇒ 00:54:36.309 Ujval Kamath: you know, how would I do that? Like I said, I played around with the, like, Bedrock has agents.
308 00:54:36.660 ⇒ 00:54:37.090 Uttam Kumaran: Yes.
309 00:54:37.400 ⇒ 00:54:40.850 Ujval Kamath: You know, it gives me a framework to sort of It…
310 00:54:41.300 ⇒ 00:54:56.190 Ujval Kamath: if I have to talk to somebody, and if I say, okay, I used Bedrock, I use Bedrock guardrails and stuff, it gives, like, a little… it’s a slightly better conversation piece than, oh, I used a bunch of open source random stuff, nobody knows what I’m talking about. Yeah. So I’ve just been playing around with a lot of AWS stuff.
311 00:54:56.320 ⇒ 00:54:57.650 Uttam Kumaran: Right. That’s kind of been my…
312 00:54:57.940 ⇒ 00:55:00.679 Ujval Kamath: How do I keep myself busy in my spare time?
313 00:55:00.680 ⇒ 00:55:01.280 Uttam Kumaran: Yeah.
314 00:55:01.550 ⇒ 00:55:15.420 Uttam Kumaran: I mean, I think, like, I mean, one, it’s been… this has been a great conversation. I think, certainly, like, we do have some clients that are at the size with some opportunities that I think your skill set could be helpful with.
315 00:55:15.610 ⇒ 00:55:17.230 Uttam Kumaran: Additionally.
316 00:55:17.340 ⇒ 00:55:25.789 Uttam Kumaran: I think you have an amazing background that could also open us up to even go to some companies that previously, with just normal
317 00:55:26.030 ⇒ 00:55:40.660 Uttam Kumaran: You know, just with the capabilities we had, maybe we weren’t able to seal the deal, but going to companies saying, hey, let’s go after some really large opportunities in the way you described, using data science, you know, maybe that’s some way to kind of collaborate.
318 00:55:40.970 ⇒ 00:55:56.020 Uttam Kumaran: You know, I think the benefit of us, and most of our folks at Brainforge started off as, like, just kind of contracting with us, and for everybody we work with, we’re sort of like, hey, just come on and, like, work on a project with us, or help us scope something, and let’s put something in front of a client.
319 00:55:57.040 ⇒ 00:56:06.749 Uttam Kumaran: You know, and that’s kind of, like, how we… everything sort of starts. But we are building, you know, a team, and so definitely also interested in… in sort of understanding, like.
320 00:56:06.760 ⇒ 00:56:14.530 Uttam Kumaran: what you’re interested in doing long-term, and whether, like, working with Brainforge and deploying, you know, great solutions to our clients, and…
321 00:56:14.530 ⇒ 00:56:28.730 Uttam Kumaran: making them and us a bunch of money in the process is interesting, you know? I think our business, as I mentioned, we’re very verticalized, but over time, I want us to go after much more complicated technical use cases.
322 00:56:28.730 ⇒ 00:56:39.299 Uttam Kumaran: So on the data side, it’s always been machine learning, data science. On the AI side, we have some use cases where we do fine-tuning, but I think we will get into more open source LLM training.
323 00:56:39.300 ⇒ 00:56:51.450 Uttam Kumaran: things like that, and I want us to go that direction, and that’s all through just having, like, an awesome team, but really the benefit of us is, like, we have a wide variety of clients, and therefore.
324 00:56:51.450 ⇒ 00:57:03.610 Uttam Kumaran: you get a wide variety of access to tons of tooling, right? And see, like, what different people are open to. Like, we work with so many different tools, and we end up with relationships with those vendors as well, because given the amount of work to do.
325 00:57:03.610 ⇒ 00:57:09.539 Uttam Kumaran: But one thing that’s starting to be really clear is that our clients just come with us with the next problem they have.
326 00:57:09.540 ⇒ 00:57:25.849 Uttam Kumaran: And they’re like, can you solve this? And so, for sure, if they wanted to build iOS apps, I’m like, we don’t do that, right? But there are some things that are more adjacent to us. Like, let me give you one example. One of our clients, they’re an e-com agency, they run ad budgets for about 80 different consumer brands.
327 00:57:25.860 ⇒ 00:57:28.410 Ujval Kamath: You can think, like, Facebook.
328 00:57:28.450 ⇒ 00:57:33.370 Uttam Kumaran: Snapchat, TikTok, like, all the major ad platforms.
329 00:57:33.840 ⇒ 00:57:39.069 Uttam Kumaran: We’re building an internal, sort of, like, Basically, like, data summarization
330 00:57:39.470 ⇒ 00:57:55.439 Uttam Kumaran: platform for them, but platform is really just chatting over a bunch of MCPs that pull data from a bunch of their platforms. We’re also starting to land some of that data and leverage AI to actually do more, like, semi-sophisticated data analysis on a daily basis that they can take
331 00:57:55.530 ⇒ 00:58:08.559 Uttam Kumaran: package and send to their clients. They want us to help them build, like, a little bit more complicated forecasting tool, where their managers, their client managers, can come in and put a bunch of forecast parameters
332 00:58:08.560 ⇒ 00:58:15.950 Uttam Kumaran: we would host some type of, you know, basically the way we scoped it is, like, we’ll post some type of Python model that sits on top of a data warehouse.
333 00:58:16.010 ⇒ 00:58:25.169 Uttam Kumaran: produces that… that forecast on a daily basis. This is something that takes them… they… one, their forecasts are extremely unsophisticated, they’re, like, just linear forecasts, like.
334 00:58:25.170 ⇒ 00:58:39.930 Uttam Kumaran: if we increase ad budgets, you’re gonna get this ROI, so it’s, like, really stupid. Second is, like, this… this is, like, gonna be a value-added offering, for their clients, and third, it’s, like, something that they don’t… they’re never gonna be able to do that without a company like us.
335 00:58:40.010 ⇒ 00:58:59.850 Uttam Kumaran: And so that’s a great opportunity, you know, there’s a clear opportunity for you and your skill set to fit in to there, for example. You know, but again, like, they’re also very interested. The reason why we took that work is it’s not just, hey, develop us this model. We’re actually developing not only the model, but the fact that our chatbots and our Slack bots
336 00:58:59.850 ⇒ 00:59:14.959 Uttam Kumaran: can pull from that, and they’re almost using it in a more ambient way to find the insights on behalf of their clients. That’s where they want to… they want to push, and we’re helping them build their internal platform for their internal, like, client managers.
337 00:59:15.190 ⇒ 00:59:32.360 Uttam Kumaran: And so that’s a mix of full-stack work, right? That is some data engineering work, there’s modeling and some data science work, there is also front-end, and they’re loving it so far, and, like, there’s a great opportunity to sort of put a scope in front of them on how we can improve that forecasting, but this is sort of, like.
338 00:59:32.460 ⇒ 00:59:49.630 Uttam Kumaran: the types of clients that we work with. We have some folks that are e-com. We also have a couple, like, multi-hundred million dollar brands that are trying to grow really, really fast, and so, okay, like, maybe they’re… yes, they have dashboards, but maybe there’s still some use cases where, okay, we do need a great
339 00:59:49.630 ⇒ 00:59:55.759 Uttam Kumaran: We do need to think about leveraging, you know, classic data science methods or advanced statistical methods.
340 00:59:55.760 ⇒ 00:59:56.950 Ujval Kamath: to address.
341 00:59:57.010 ⇒ 01:00:07.349 Uttam Kumaran: But, you know, also, it’s like, just thinking about the things that, you know, you’ve done allows us to even go into other industries, you know, where they’re having the same problems, and so…
342 01:00:07.350 ⇒ 01:00:21.970 Uttam Kumaran: I mean, I’m certainly interested in potentially working together, and would love to even connect you with some more people on my team, just to… so you could chat and learn more, and hear a bit more about, like, what it’s like day-to-day on a project, and things like that, if you’re okay with that.
343 01:00:22.730 ⇒ 01:00:25.929 Ujval Kamath: Yeah, I’d definitely like to talk more. It seems pretty interesting, and certainly, like.
344 01:00:26.150 ⇒ 01:00:35.580 Ujval Kamath: I think I’m with you, like, I’ve worked in, sort of, the customer-facing and large companies, and it can be good, and you know the bad as well, I think you’ve been in that situation.
345 01:00:35.580 ⇒ 01:00:44.409 Uttam Kumaran: Internally, yeah, it’s tough, it’s really tough, and you have to become a people manager to get any, like, power. You have to be, like, a Roman Empire, like, you know, it’s like…
346 01:00:44.410 ⇒ 01:00:53.300 Ujval Kamath: Exactly. So, I mean, the… what you’re describing to me, like, how your setup sounds, certainly a lot more, appealing than… than what that was, so…
347 01:00:53.300 ⇒ 01:00:58.300 Uttam Kumaran: Yeah, and there’s also really practical matters. One, it’s like, our…
348 01:00:58.530 ⇒ 01:01:02.470 Uttam Kumaran: Our, like, the leadership here in our company is very, like.
349 01:01:02.580 ⇒ 01:01:11.399 Uttam Kumaran: it’s a pretty chill environment. Like, we are a startup by the fact that we’ve only been in business a few years, but there’s no startup really about this, like…
350 01:01:11.510 ⇒ 01:01:26.810 Uttam Kumaran: It’s not… it’s not chaotic at all. There’s no, like, chaos. Like, we deliver for clients, and yes, our clients may be chaotic, but our job is that their engagement with us is not. Like, we are getting more and more formalized and, predictable
351 01:01:26.900 ⇒ 01:01:42.470 Uttam Kumaran: As we grow. We’re also getting ability to take swings at companies that a 2- or 3-year-old company should never have the opportunity to do so, because of the great testimonials we have, because of our network, and sort of our ability… we do a lot of marketing and sales, and so…
352 01:01:42.470 ⇒ 01:01:52.140 Uttam Kumaran: we’re getting a chance at working at those problems, and third is, like, a lot of times, companies have just never worked… like, typically when they work for the service, they get talked to an account manager, a project manager.
353 01:01:52.140 ⇒ 01:01:52.890 Ujval Kamath: Oh, yeah, oh yeah.
354 01:01:52.890 ⇒ 01:02:12.649 Uttam Kumaran: engineering managers, like, all these people that are, like, useless. I’m telling you, like, our engineers at our company, they’re all client-facing. Like, they’re all, like, able to engage and put together decks, memos, but also do the work, and we’ve brought them… brought everyone to that, because clients love working with the people that are actually doing the work.
355 01:02:12.700 ⇒ 01:02:26.990 Uttam Kumaran: Versus… and also for me, like, I don’t want to… we try to hire project managers, and it’s, like, so painful. And I was a product manager, you know, at one point, and a lot of that, we’re using AI to just smooth out the gaps between handoffs and…
356 01:02:27.090 ⇒ 01:02:46.019 Uttam Kumaran: I don’t know, we’re trying this, like, new way of running this type of company, and so far the results have been very, very good, and we’ve… but we’ve never, we’ve never sacrificed the quality of the work. Like, we don’t do staff augmentation, we don’t do, like, dev shop style stuff at all.
357 01:02:46.160 ⇒ 01:02:59.450 Uttam Kumaran: like, most of our folks are mid-level or senior, and we’re delivering, like, really, really great, quick outcomes for clients, and they expect us to be able to explain that to an executive audience. And so, like, that’s the kind of stuff
358 01:02:59.660 ⇒ 01:03:15.429 Uttam Kumaran: that we do, and I think we’re… we’re somewhat unique in, like, our positioning, and we made a really conscious effort to not get relegated as just, like, the engine data team, right? We’re more of, like, a partner to a strategic, like, business opportunity.
359 01:03:15.680 ⇒ 01:03:25.640 Uttam Kumaran: And that’s how we’re framing everything for the client as well. So, yeah, I mean, I… I… I usually try to think about, like, what it would be like to work here, and…
360 01:03:25.640 ⇒ 01:03:38.380 Uttam Kumaran: I think it’s been getting… my expectations are very high, but it’s been getting better and better and better. Like, I think all of our team has been really enjoying getting a wide variety of experience, a wide variety of clients.
361 01:03:38.780 ⇒ 01:03:58.110 Uttam Kumaran: We’ve… people are growing, like, really, really fast, which is great, and we’re using AI for every part of the manufacturing process of our service. You know, so let… not only are we doing AI for clients, our analysts are all using Cursor for all their analysis, using MCP to query directly to the data warehouse.
362 01:03:58.450 ⇒ 01:04:05.519 Uttam Kumaran: We’re using AI for deck creation, for, like, understanding, like, certain industries, like competitive analysis.
363 01:04:05.540 ⇒ 01:04:15.970 Uttam Kumaran: But, like, every part of the manufacturing process of data we’re trying, whether it’s also just using AI to help you write dbt models faster, or, like, you don’t have to go to Snowflake docs to, like, look something up, right?
364 01:04:15.970 ⇒ 01:04:26.740 Uttam Kumaran: in every which way we’re trying to do that, because ultimately, we’re developing a product, we’re developing a better service faster, which means we’re actually, like, a premium. It’s not like we…
365 01:04:26.740 ⇒ 01:04:34.300 Uttam Kumaran: And the other thing we’ve been doing is trying to move away from, like, the billable hour. Most of our stuff is fixed scope objectives.
366 01:04:34.300 ⇒ 01:04:47.790 Uttam Kumaran: But not in, like, a, hey, come do this and leave. It’s, like, long-term, like, almost like you’re subscribing to, like, a team-type engagement. And people love that because they’re like, here’s just a budget, like, do as much as you can.
367 01:04:47.800 ⇒ 01:04:51.390 Uttam Kumaran: you know, within that. And so, I feel like we’ve been able to push at these, like.
368 01:04:51.860 ⇒ 01:05:11.510 Uttam Kumaran: kind of hairy topics in consulting right now, whereas the big consultancies are just, like, firing everybody, because they’re, like, realizing that half the folks, that their job can be totally done with AI, and that they bill hourly, so then they’re like, what? I don’t know how we reconcile these. So, that’s a little bit about, like, how we’re thinking about things.
369 01:05:13.260 ⇒ 01:05:16.149 Ujval Kamath: Yeah, I mean, it sounds super interesting. I’d love to keep talking.
370 01:05:16.150 ⇒ 01:05:16.960 Uttam Kumaran: Fuck.
371 01:05:17.990 ⇒ 01:05:31.639 Uttam Kumaran: Yeah, let me, so let me talk to my team. Rico on my team will sort of connect you with a few people. Yeah, I mean, feel… I would love for you to just grill them on… on what it’s like, and types of projects, and what it’s like working with me, and, like, their time at Brainforge.
372 01:05:31.670 ⇒ 01:05:38.569 Uttam Kumaran: And, like, yeah, definitely, you know, any questions you have for me, I’ll send you my phone number as well. Any questions you have.
373 01:05:38.640 ⇒ 01:05:41.380 Uttam Kumaran: Along the way, more than happy to…
374 01:05:41.570 ⇒ 01:05:55.820 Uttam Kumaran: to, to answer those. We’re pretty open book, so… and we’re all… we’re all data people, so we’re all… we’ve all done a lot in the industry, so, you know, yeah, I’m really looking forward to, sort of, next steps.
375 01:05:56.870 ⇒ 01:05:57.799 Ujval Kamath: Yeah, me too.
376 01:05:58.400 ⇒ 01:06:03.130 Uttam Kumaran: Awesome. Okay, well, thank you so much. I know we went over, so I appreciate the time, and sorry again for being… That’s fine.
377 01:06:03.660 ⇒ 01:06:07.550 Ujval Kamath: Oh, well, thanks, thanks, Utemp, for talking to me, it was great. I really enjoyed the conversation.
378 01:06:07.980 ⇒ 01:06:09.419 Uttam Kumaran: Okay, I’ll talk to you soon.
379 01:06:09.570 ⇒ 01:06:10.530 Ujval Kamath: Yeah, bye.