Meeting Title: Brainforge Interview w- Sam Date: 2026-03-05 Meeting participants: Samuel Roberts, joaquin, Kaela Gallagher
WEBVTT
1 00:00:20.020 ⇒ 00:00:21.710 joaquin: Hi, can you hear me?
2 00:00:22.370 ⇒ 00:00:23.480 Samuel Roberts: Yes, I can now.
3 00:00:24.130 ⇒ 00:00:25.090 Samuel Roberts: You hear me?
4 00:00:25.260 ⇒ 00:00:26.410 joaquin: Yeah, I can hear you.
5 00:00:26.700 ⇒ 00:00:27.919 Samuel Roberts: Excellent. How are you?
6 00:00:28.120 ⇒ 00:00:29.599 joaquin: Yeah, I’m doing great. How you doing?
7 00:00:30.050 ⇒ 00:00:33.929 Samuel Roberts: Good, good. Sam, yes, yeah, thanks for taking the time.
8 00:00:33.930 ⇒ 00:00:34.880 joaquin: Yeah, for sure.
9 00:00:35.950 ⇒ 00:00:41.910 Samuel Roberts: Oh, okay, yeah. Kayla’s gonna be here, she’s a recruiting manager, but she’ll just be camera off, hanging out.
10 00:00:43.800 ⇒ 00:00:51.599 Samuel Roberts: Yeah, so, just brief intro of myself. My name’s Sam Roberts, I’m the AI tech lead here at Brainforge.
11 00:00:51.830 ⇒ 00:00:59.619 Samuel Roberts: I think we’ll just start… you can give an intro yourself, I’ve got some questions, I’ll leave time for you to ask questions, and that’ll be basically it.
12 00:00:59.820 ⇒ 00:01:03.049 joaquin: Sure, sound good. That’s cool. Oh, sorry.
13 00:01:03.050 ⇒ 00:01:04.170 Samuel Roberts: You’re fine, you’re fine, go ahead.
14 00:01:04.489 ⇒ 00:01:15.249 joaquin: Yeah, just kicking us off, my name is Joaquin. I’ve been a software engineer for, you know, a little over a decade, mostly working at early-stage and high-growth startups.
15 00:01:15.249 ⇒ 00:01:34.729 joaquin: more recently was at a company called, LeafLink, which is a marketplace for cannabis companies, so, but business-to-business, so think of, like, Amazon, but for cannabis. And one of the problems cannabis companies face in the industry is that, because it’s illegal on the federal level still.
16 00:01:34.879 ⇒ 00:01:54.669 joaquin: they’re not afforded the traditional payment methods, so they can’t use Visa, MasterCard, or bank with your traditional banking institutions like JP Morgan or Bank of America. And so, more recently, have been focused on delivering payment solutions and banking solutions for our cannabis customers.
17 00:01:55.269 ⇒ 00:01:56.969 Samuel Roberts: Adam, pass it over to you.
18 00:01:59.470 ⇒ 00:02:12.129 Samuel Roberts: Yeah, so I just, got a couple questions here. So, what kind of work… is it… have you been doing AI-focused work there at all, or how has it been, I’m just curious where to… where to go with this interview, yeah.
19 00:02:12.130 ⇒ 00:02:29.430 joaquin: Yeah, sure. So, I mean, we use AI in our workflow, so, of course, we’re really, built in with Cursor, GBT, and then have done a little model training, and I can get into that, deeper if you like.
20 00:02:29.940 ⇒ 00:02:34.270 Samuel Roberts: Okay, so have you, built any, like, LLM-based features or anything?
21 00:02:34.270 ⇒ 00:02:38.989 joaquin: I haven’t built any LLM-based features, I’m looking to get deeper into that, but did.
22 00:02:38.990 ⇒ 00:02:39.450 Samuel Roberts: Okay, cool.
23 00:02:39.450 ⇒ 00:02:42.279 joaquin: Model training for, image recognition.
24 00:02:42.950 ⇒ 00:02:43.520 Samuel Roberts: Okay.
25 00:02:44.220 ⇒ 00:02:45.160 Samuel Roberts: Cool, cool, cool.
26 00:02:45.350 ⇒ 00:02:48.609 Samuel Roberts: Sorry, a lot of my questions are, like, focused on that sort of stuff, so…
27 00:02:49.630 ⇒ 00:02:52.449 Samuel Roberts: You’re good, you’re good! Let’s, let’s do, like, you know…
28 00:02:53.010 ⇒ 00:02:56.440 Samuel Roberts: question that we can, we can twist a little bit. So, you know.
29 00:02:56.640 ⇒ 00:03:07.420 Samuel Roberts: we, you know, have client work, so we have different stakeholders, different projects. We also have some internal work, which we’ll talk about. But, you know, a lot of non-technical people that we’re interfacing with.
30 00:03:08.700 ⇒ 00:03:20.700 Samuel Roberts: whether that be clients or other people at the company, and so sometimes there’s a problem explaining the limitations of some of these things, especially, like, LLM tools and, how they’re maybe different than some things they might have expected in the past, but they see tons of stuff
31 00:03:20.700 ⇒ 00:03:33.390 Samuel Roberts: out there, and they’re always excited about it. Right. So I’m wondering, how would you think about or go about explaining the limitations of, say, an LLM tool, or an example that you have in your history to a non-technical stakeholder?
32 00:03:37.170 ⇒ 00:03:45.370 joaquin: Cool, so I’m gonna think about, specifically, this wasn’t, LLM, but this is model training for,
33 00:03:45.780 ⇒ 00:04:00.380 joaquin: for one of the features we offered in banking, and so I kind of want to start off with context, which is typically not technical. So, one of the solutions we offered was banking, and alongside that comes with the ability to deposit your checks, right?
34 00:04:01.190 ⇒ 00:04:12.100 joaquin: And so, users can take a picture of their check and upload it to our backend, and our backend would then extract the check information programmatically using OCR.
35 00:04:12.280 ⇒ 00:04:13.090 joaquin: Right?
36 00:04:13.450 ⇒ 00:04:17.799 joaquin: M… And this system was failing about 40% of the time.
37 00:04:18.820 ⇒ 00:04:25.920 joaquin: And that’s because it didn’t have… when it was initially deployed, it didn’t have the initial data to train on.
38 00:04:26.520 ⇒ 00:04:34.459 joaquin: And so that’s kind of the constraint there. In addition to that, another constraint is that there’s variability to handwritten amounts.
39 00:04:35.150 ⇒ 00:04:51.669 joaquin: So you have, like, so a customer can write 500 in the handwritten amount for a check, and the 5 might be, misinterpreted as an S for the 0 might be misinterpreted as an O. And so these are kind of the constraints of our current system.
40 00:04:53.790 ⇒ 00:04:59.849 joaquin: The solution was, now that we have been up and running for a while, we have all this proprietary check data.
41 00:04:59.980 ⇒ 00:05:08.409 joaquin: And so we can retrain our model to adapt to… to adapt and make the corrections, given this new data set.
42 00:05:08.690 ⇒ 00:05:12.790 joaquin: And so, you know, in this example, we improved the system by, you know, double.
43 00:05:13.060 ⇒ 00:05:27.640 joaquin: from 40% success rate to, like, 80% success rate. Reason being, handwritten amounts was, the biggest constraint or variability there, and we didn’t have enough of a data set to really improve handwritten values.
44 00:05:28.610 ⇒ 00:05:29.210 Samuel Roberts: Cool.
45 00:05:29.370 ⇒ 00:05:31.650 Samuel Roberts: Okay, yeah, that’s great.
46 00:05:33.830 ⇒ 00:05:40.540 Samuel Roberts: Yeah, excellent, okay. Like, I’m looking at my, you know, AILM questions and trying to tweak them a little bit here, but… Yeah.
47 00:05:40.540 ⇒ 00:05:47.149 joaquin: Apologies if there was a misunderstanding that I was going for, like, an AI role here.
48 00:05:47.150 ⇒ 00:05:47.620 Samuel Roberts: Yeah.
49 00:05:47.620 ⇒ 00:05:51.280 joaquin: Yeah, frankly, I’m not, like, .
50 00:05:51.370 ⇒ 00:05:54.219 Samuel Roberts: I’m not deep into LLM right now. Sure.
51 00:05:54.640 ⇒ 00:05:58.879 joaquin: I’m looking… I’m taking this course called, Deep Atlas.
52 00:05:59.230 ⇒ 00:06:07.190 joaquin: which kind of teaches you how to build LLMs from the ground up, and so, you know, it is my…
53 00:06:07.310 ⇒ 00:06:18.840 joaquin: goal to get into the industry on the lower level, but, you know, I think I was, I was under the impression, that this was more of, like, an engineering
54 00:06:19.130 ⇒ 00:06:26.820 joaquin: opportunity, not specifically from a low-level, AI or LLM.
55 00:06:27.350 ⇒ 00:06:42.100 Samuel Roberts: Yeah, yeah, no, I mean, it’s… I can give you a little bit more context there. So, I mean, we’re, you know, the AI automations team is kind of one of the two main engineering teams, the other side being the data side. So, you know, they’re doing ETLs and pipelines and modeling and all that stuff, and we’re kind of doing…
56 00:06:42.380 ⇒ 00:06:56.419 Samuel Roberts: whatever else. So that includes internal tooling, the platform that we use that ingests our meetings and all that sort of stuff. That includes client work from, you know, RAG-assisted chatbots to just automating some AI flows that people are already doing in their
57 00:06:56.420 ⇒ 00:07:05.109 Samuel Roberts: Claude, to, you know, pulling in data from MCPs and helping people under… we kind of cover a lot of different stuff, so there’s definitely a strong focus on AI.
58 00:07:05.110 ⇒ 00:07:05.490 joaquin: Right.
59 00:07:05.490 ⇒ 00:07:10.990 Samuel Roberts: tooling and using that. We’re not at a really low level in terms of that. We’re not, you know, training models or anything yet.
60 00:07:10.990 ⇒ 00:07:11.460 joaquin: Okay.
61 00:07:11.460 ⇒ 00:07:16.890 Samuel Roberts: But we’re kind of… internally, we’re using all the newest stuff and trying to test things out and see what…
62 00:07:16.990 ⇒ 00:07:27.779 Samuel Roberts: tools and models and frameworks, like, work and will work long-term. And for client work, we’re kind of using probably, like, maybe not the most, most recent stuff, but we’re kind of staying on top of it, and…
63 00:07:27.780 ⇒ 00:07:41.069 Samuel Roberts: and doing that. So, yeah, I mean, I… a lot of the questions are kind of, like, if people have already been building, you know, with models and things, how they’re… how they’re doing it. So that’s sort of why I was just, like, the questions for the AI engineer interviews are a little bit skewed that way, but.
64 00:07:41.070 ⇒ 00:07:42.219 joaquin: No, for sure.
65 00:07:42.220 ⇒ 00:07:42.940 Samuel Roberts: Yeah.
66 00:07:42.940 ⇒ 00:07:45.949 joaquin: I was totally getting that, that vibe. I mean,
67 00:07:47.930 ⇒ 00:07:52.690 joaquin: Is your background in AI and ML low level?
68 00:07:52.900 ⇒ 00:08:06.080 Samuel Roberts: No, no, so I come from a startup, or product background. Obviously the last several years has been a lot more AI stuff, but, I was more focused on building, you know, a couple
69 00:08:06.490 ⇒ 00:08:12.650 Samuel Roberts: customer-facing things, a couple more B2B things, really web tech stuff.
70 00:08:13.760 ⇒ 00:08:22.969 Samuel Roberts: And so I’m… I’m still learning a lot of this stuff, because it’s changing all the time too, but, yeah, I don’t have a strong, like, ML background pre, you know.
71 00:08:23.580 ⇒ 00:08:25.480 Samuel Roberts: LLM’s kind of jumping on the scene.
72 00:08:25.480 ⇒ 00:08:26.660 joaquin: Okay.
73 00:08:27.540 ⇒ 00:08:34.789 joaquin: Yeah. I wanted to add, maybe it… and I’m not sure what the direction is from here, because it seemed like a lot of,
74 00:08:34.840 ⇒ 00:08:45.920 joaquin: your… the questions you did want to ask were more AI and ML focused, and that’s totally true, but maybe, you know, once I’m done with the program, maybe it might make sense to kind of…
75 00:08:45.940 ⇒ 00:08:55.049 joaquin: reconnect, especially if the role is more low… closer to low level, or more AI ML familiar.
76 00:08:55.690 ⇒ 00:09:04.010 joaquin: Does… does that make sense? Or… or is it, like, still a consideration that you are hiring for, like, software engineers, for example?
77 00:09:04.010 ⇒ 00:09:10.990 Samuel Roberts: I mean, yeah, we’re definitely looking for, you know, software engineers, but we’re all kind of figuring out a lot of the, you know.
78 00:09:11.650 ⇒ 00:09:14.630 Samuel Roberts: AI tooling, AI, you know, stuff.
79 00:09:15.410 ⇒ 00:09:23.919 Samuel Roberts: If you’re interested in that, like, that’s, you know, there’s definitely… we’re doing software engineering, we’re mostly, like, a TypeScript-Python kind of split, depending on what we’re doing, sort of shop.
80 00:09:24.820 ⇒ 00:09:30.390 Samuel Roberts: Most of our projects are… AI-related or automation, I guess, related in some form.
81 00:09:30.390 ⇒ 00:09:31.430 joaquin: Right.
82 00:09:31.790 ⇒ 00:09:41.220 Samuel Roberts: if that’s interesting, like, you know, I, I, you know, but if it’s not, I also totally understand if you want to, you know, focus on certain other things. So, yeah, you tell me, I guess, if that’s still something you’re.
83 00:09:42.170 ⇒ 00:09:47.309 joaquin: Yeah, so I think, like, my interest in Brainforge is, like, you’re consulting across various,
84 00:09:47.630 ⇒ 00:09:55.110 joaquin: Different companies and industries, and, you’re probably gonna have the most experience with breadth in terms of
85 00:09:55.220 ⇒ 00:10:06.970 joaquin: AI and the applications to the various problems that people are facing, and so, like, I thought it would be, like, a great experience to, you know, given that I don’t have a deep experience,
86 00:10:07.570 ⇒ 00:10:13.500 joaquin: a good experience to kind of, build on that. So, that’s kind of, like, what drew my interest.
87 00:10:13.730 ⇒ 00:10:29.589 Samuel Roberts: Yeah, and I think… I think it is, you know, like, I… I had some AI experience, some building some AI products, or AI-infused products, I guess. But I’ve definitely done a lot more, you know, since… since joining, and obviously things changed, and so there’s always new things. Right.
88 00:10:29.910 ⇒ 00:10:35.250 Samuel Roberts: But yeah, so I mean, if you’re interested in that kind of a… sorry, my cat keeps jumping on my lap.
89 00:10:35.960 ⇒ 00:10:45.109 Samuel Roberts: Yeah, I mean, I think, you know, strong engineering backgrounds are definitely what we’re looking for. I guess the AI side of it is more interest and ability to learn, more than anything.
90 00:10:45.110 ⇒ 00:10:45.680 joaquin: I’m…
91 00:10:46.480 ⇒ 00:10:50.159 Samuel Roberts: You know, there’s not… you’re not gonna have 10 years of experience doing, you know.
92 00:10:50.790 ⇒ 00:10:59.689 Samuel Roberts: building RAG systems or something, probably, at least in the broad sense here. So yeah, I mean, if you’re… if this is the type of role that’s intriguing to you.
93 00:11:00.990 ⇒ 00:11:10.779 Samuel Roberts: I’m happy to keep going, but if you’re definitely, like, I want to get a little more under my belt before, I think this is a good place to do that, but also, I understand if you’re not looking to, like, jump right into that, so…
94 00:11:10.780 ⇒ 00:11:25.440 joaquin: Yeah, it sounds like there’s still alignment from what I’m hearing. There might be, like, a few logistics things about me finishing up the program, should everything work out with Brainforge. But yeah, it sounds like we can continue on.
95 00:11:25.930 ⇒ 00:11:30.930 Samuel Roberts: Alright, cool, cool. Yeah…
96 00:11:31.880 ⇒ 00:11:46.320 Samuel Roberts: I guess I… I’m curious to learn a bit more about, your experience, then. So, LeafLink, I’m looking, I have your… your LinkedIn pulled up here. Yeah. So are you mostly Python?
97 00:11:47.040 ⇒ 00:11:59.160 joaquin: Yeah, so we were mostly a Python shop. We went on to acquire some banking technology, and that was written in Java. Okay. So experience with Python and Java on the front end was,
98 00:11:59.160 ⇒ 00:12:09.569 joaquin: Vue.js and TypeScript, so a little bit familiar with that, but I wouldn’t consider myself, like, a front-end architect. Sure, totally.
99 00:12:10.070 ⇒ 00:12:10.570 joaquin: Yeah.
100 00:12:10.570 ⇒ 00:12:15.900 Samuel Roberts: Okay, cool. And did I see on here YC Winter 2019?
101 00:12:15.900 ⇒ 00:12:16.780 joaquin: Yeah, yeah.
102 00:12:16.780 ⇒ 00:12:19.059 Samuel Roberts: Were you… were you out in California then with the…
103 00:12:19.060 ⇒ 00:12:19.880 joaquin: Yep, yeah.
104 00:12:19.880 ⇒ 00:12:21.400 Samuel Roberts: I was in that class, actually.
105 00:12:21.400 ⇒ 00:12:22.249 joaquin: Oh, no way!
106 00:12:22.250 ⇒ 00:12:30.480 Samuel Roberts: Yeah, when I saw that, I was like, what? And then I went and scrolled down, I remember… I remember… I don’t know if I remember you necessarily, but I remember some people from Doorport, there were, like, two founders, I think, at that time.
107 00:12:30.480 ⇒ 00:12:31.310 joaquin: Yeah, two founders.
108 00:12:31.310 ⇒ 00:12:32.070 Samuel Roberts: Yeah, yeah, yeah, okay.
109 00:12:32.070 ⇒ 00:12:33.070 joaquin: didn’t benefit.
110 00:12:33.070 ⇒ 00:12:41.140 Samuel Roberts: Yeah, yeah, yeah, okay, they were… I don’t know if they were in our day or not, but it’s been a long time at this point, but I saw Winter 19, and yeah, I was… I was… my background a little bit more here, but…
111 00:12:41.550 ⇒ 00:12:52.849 Samuel Roberts: I was at a hair care company that I started with a friend, and so… not tech, kind of different, not really typical YC, but yeah, I was there for that class, so that’s kind of funny.
112 00:12:52.850 ⇒ 00:12:54.130 joaquin: That’s crazy.
113 00:12:54.130 ⇒ 00:12:54.630 Samuel Roberts: Yeah.
114 00:12:54.630 ⇒ 00:12:57.519 joaquin: Wow, and you were in that, you were in that class, yeah.
115 00:12:57.520 ⇒ 00:12:58.599 Samuel Roberts: Yeah, yeah.
116 00:12:58.900 ⇒ 00:13:17.250 joaquin: Oh my god. Yeah, so yeah, Doorport. Yeah, so I did the whole… I was… I wasn’t, like, technically a founder, but I was able to go to, like, demo day, you know, sat in some of the investor meetings, and, raised some capital, and ended up selling the company, and so,
117 00:13:17.710 ⇒ 00:13:19.780 joaquin: So, so good experience, good learning.
118 00:13:19.780 ⇒ 00:13:20.390 Samuel Roberts: Yeah, cool, okay.
119 00:13:20.440 ⇒ 00:13:22.189 joaquin: Wild that you were… you were…
120 00:13:22.190 ⇒ 00:13:30.470 Samuel Roberts: Yeah, when I saw it, I was like, you know, I see YC listed sometimes, but I very rarely am I seeing the exact one I, that class I was in. Cool, okay.
121 00:13:31.410 ⇒ 00:13:36.230 Samuel Roberts: Yeah, I guess… I’m trying to think where to jump from here.
122 00:13:37.870 ⇒ 00:13:53.770 Samuel Roberts: So let’s talk about, you know, working on a team. You know, I don’t know where exactly… I have your LinkedIn here, but I’m not sure where exactly how big of teams you’ve worked on, where in sort of leadership versus not, like, can you just give me a kind of rundown of that, at least in the last few years?
123 00:13:53.950 ⇒ 00:13:59.949 joaquin: Yeah, sure. So… Let’s start off with…
124 00:14:00.120 ⇒ 00:14:02.760 joaquin: maybe we could just talk about my journey at LeafLink, where I kind of…
125 00:14:02.760 ⇒ 00:14:03.210 Samuel Roberts: True.
126 00:14:03.210 ⇒ 00:14:22.350 joaquin: as an individual contributor, I kind of rose through the ranks, and then, you know, when the time came, and we launched banking, I kind of led the backend team there for the organization’s largest pod, which was a team of 9 engineers.
127 00:14:22.380 ⇒ 00:14:28.149 joaquin: And this was a zero-to-one project, with the stakeholders being, like.
128 00:14:28.170 ⇒ 00:14:48.009 joaquin: those on the executive level, because this is a new project. So, like, my stakeholder would be, like, the CTO and the chief business officer, and then I’ve kind of fell into a little bit more of a manager role on the banking team, as opposed to, you know, focusing on the nitty-gritty, low-level work.
129 00:14:48.180 ⇒ 00:14:51.599 joaquin: So that’s, like, my growth story at LeafLink.
130 00:14:52.080 ⇒ 00:14:52.690 Samuel Roberts: Okay.
131 00:14:53.430 ⇒ 00:15:04.049 Samuel Roberts: So, I guess that kind of then informs, like, what you want to get into, kind of, this… this world a little bit. You said, kind of, lower level stuff, like, where does… where does you see yourself in the next…
132 00:15:04.200 ⇒ 00:15:05.879 Samuel Roberts: You know, several years.
133 00:15:06.250 ⇒ 00:15:13.280 joaquin: Yeah, so, I mean… the… the nature of, kind of, like, this management role that I played
134 00:15:13.540 ⇒ 00:15:18.300 joaquin: For the banking team was more of… circumstantial.
135 00:15:18.560 ⇒ 00:15:19.340 Samuel Roberts: Okay.
136 00:15:21.300 ⇒ 00:15:40.300 joaquin: one, our team didn’t have a lead, and two, it didn’t have, like… when it first started, it didn’t have a product manager or engineering manager. And so what the team did was lean into me for, like, the organizational, and being the lead eyes on between the team and, the executive suite.
137 00:15:40.730 ⇒ 00:15:50.449 joaquin: And so, that… that was a role I was encouraged to be in, but I kind of want to focus more on, the IC path as now.
138 00:15:50.580 ⇒ 00:15:58.740 joaquin: And if, like, management is something that I could grow into, and that’s what the team would want of me.
139 00:15:59.080 ⇒ 00:16:11.820 joaquin: that I’m open to, having had the experience, but I think with AI, there’s, like, so much new to learn, from a technical perspective, so it kind of is… is… the IC path is refreshing again.
140 00:16:12.200 ⇒ 00:16:22.630 Samuel Roberts: I understand that, yeah. So how… how big… you mentioned 9 people on the pod, how big is LeafLink? Like, I’m just trying to understand, like, you know, Doorport, more startup-y, I’m wondering where you’re, you know.
141 00:16:23.570 ⇒ 00:16:28.740 Samuel Roberts: what you’re looking for, I guess, in terms of that, you know, size-wise, or… Culture-wise.
142 00:16:28.960 ⇒ 00:16:33.170 joaquin: Yeah, so… we…
143 00:16:33.550 ⇒ 00:16:40.800 joaquin: In its height, the engineering organization was 80 engineers, and I was there from, like, early on to, like, that high-growth peak.
144 00:16:41.770 ⇒ 00:16:50.879 joaquin: The company then had several reductions in force, and so, like, the size of the engineering team was whittled down to, like.
145 00:16:51.110 ⇒ 00:16:52.680 joaquin: 8 engineers.
146 00:16:53.130 ⇒ 00:16:53.910 Samuel Roberts: Wow.
147 00:16:53.910 ⇒ 00:17:03.490 joaquin: Yeah. In terms of, like, what I’m looking for in terms of size, nothing particular. I’m not looking for… I know I’m not looking for, like, big tech, big organization.
148 00:17:03.490 ⇒ 00:17:04.180 Samuel Roberts: Sure.
149 00:17:04.180 ⇒ 00:17:20.879 joaquin: I feel like I can deliver the most meaningful impact, at smaller companies and smaller teams, and that’s kind of what’s brought me towards the early-stage startup life throughout my entire career. And so that’s what I have, like, a particular affinity for.
150 00:17:21.150 ⇒ 00:17:29.579 Samuel Roberts: Cool. Okay, good. Yeah, that… okay, yeah, we’re definitely, more startup-y, you know, it’s a little different, because it’s, you know, more consultant.
151 00:17:29.750 ⇒ 00:17:35.200 Samuel Roberts: consultancy kind of work, but definitely have that… has that ethos, I think. Right.
152 00:17:35.510 ⇒ 00:17:44.019 Samuel Roberts: Cool, alright, so, we’re about halfway. Are there other questions you have for me, about Brain Forge, about the work, whatever I can answer for you?
153 00:17:44.020 ⇒ 00:17:48.040 joaquin: Yeah, definitely. I kind of wanted to learn a little bit more about your background.
154 00:17:48.040 ⇒ 00:17:48.550 Samuel Roberts: Sure.
155 00:17:48.550 ⇒ 00:17:56.769 joaquin: I know that you went to YC, you’re a founder, I got the impression that you’re mostly in product, maybe you can speak to a little bit of that.
156 00:17:57.160 ⇒ 00:18:06.350 Samuel Roberts: Yeah, so, you know, very high level, I was a mechanical engineer by training, got into startups kind of right out of school,
157 00:18:06.870 ⇒ 00:18:24.170 Samuel Roberts: I did a program called Venture for America. So, after graduating, I was kind of… I picked a company to work with here in Cleveland. I went to Case Western here in Cleveland, I’m still here. I also started a company at the same time with a friend right out of school, so kind of doing these two things, and at that company, I learned a ton of,
158 00:18:24.560 ⇒ 00:18:42.360 Samuel Roberts: well, web dev, I would say, but it was really… it was a company that made these little sensors for doing energy audits of buildings, and the guy who invent… who built these chips, he programmed them, you know, all the low-power chips, and he built the whole stack. The backend was all Node, which at that point, you know, I knew what JavaScript was, but didn’t really understand, like.
159 00:18:42.590 ⇒ 00:18:44.380 Samuel Roberts: Browser vs server kind of stuff.
160 00:18:44.570 ⇒ 00:18:45.280 joaquin: Yeah.
161 00:18:45.280 ⇒ 00:18:59.199 Samuel Roberts: So I was thrown into this world of JavaScript on the server without really knowing what I was doing, figured it out, you know, then left that company to work on mine, and we went kind of full-stack JavaScript, so I kind of really cut my teeth there.
162 00:18:59.370 ⇒ 00:19:04.510 Samuel Roberts: Did some freelancing. While I was freelancing, ended up helping out this friend at the hair care company.
163 00:19:04.680 ⇒ 00:19:13.989 Samuel Roberts: So I kind of put down the keyboard for a few years, and literally made hair products, for a while in my kitchen, and then grew that. We did…
164 00:19:14.420 ⇒ 00:19:24.929 Samuel Roberts: I think the year I left, it was, like, $5 million in revenue. We had a warehouse here in Cleveland making hair products, fulfilling orders, co-packing with other people to make shampoos and stuff,
165 00:19:25.070 ⇒ 00:19:45.059 Samuel Roberts: realized I didn’t want to be in hair care. That was not really… you know, it was originally helping out a friend and became a 5-year thing, and so I left that, right around the pandemic, took a little time off, ended up moving to London, and did another startup there with a guy I met there, called Moot, and that was, like, a collaborative workspace, so this is, like, 2021, 2022.
166 00:19:46.010 ⇒ 00:19:59.109 Samuel Roberts: crowded industry, he had some money raised from a very friendly investor for another project he had worked on, so we were really just burning through the rest of his money, trying to see if we could get something to spark. And it was a really interesting product, it was really cool to build, it was very front-end heavy.
167 00:19:59.120 ⇒ 00:20:11.219 Samuel Roberts: The backend was literally just, like, Superbase as a CRUD API, and then everything was, collaborative, so there was real-time syncing, and Google Doc-style notes, and a whiteboard, and all this other stuff. So that was great, because I…
168 00:20:11.510 ⇒ 00:20:13.760 Samuel Roberts: Was coming from older tech.
169 00:20:13.760 ⇒ 00:20:33.040 Samuel Roberts: you know, JavaScript changed, TypeScript became a thing, React became a thing, I was using Angular back in the day, so a lot of that kind of shifted, and I was able to kind of get back up to speed. Worked on that for a few years, moved back to the States, and was looking for kind of my next thing. I’m older now than I was when I first got into startups, so I was looking for something maybe a little.
170 00:20:33.040 ⇒ 00:20:33.440 joaquin: push more.
171 00:20:33.440 ⇒ 00:20:40.600 Samuel Roberts: stable, you know, I’ve got a kid and a house and a wife, and so, this is a… this has been a good place because it’s giving me that kind of,
172 00:20:41.030 ⇒ 00:20:51.799 Samuel Roberts: you know, scratching that itch of working on cool things, going 0 to 1 with a lot of stuff for clients, but I’m not starting a company and all the things that that entails, so I’ve been enjoying, my time here, where it…
173 00:20:51.800 ⇒ 00:21:01.929 Samuel Roberts: It’s a little different than what I’m used to, because I’m used to, you know, kind of one product, one project, and being on a smaller team of me and maybe one other engineer and the designer, you know, as founders kind of thing.
174 00:21:01.950 ⇒ 00:21:06.280 Samuel Roberts: So that’s been… that’s sort of my journey to get here. It’s a little…
175 00:21:06.280 ⇒ 00:21:07.060 joaquin: That’s awesome.
176 00:21:07.060 ⇒ 00:21:08.480 Samuel Roberts: Little weird,
177 00:21:08.480 ⇒ 00:21:18.999 joaquin: Yeah, that’s… I think it’s… it’s… for me, it’s pretty relatable, because, like, one, obviously, YC Winter 19, but then also you have, like, extensive experience as, like, early, early employee,
178 00:21:19.000 ⇒ 00:21:19.430 Samuel Roberts: Yeah.
179 00:21:19.430 ⇒ 00:21:30.229 joaquin: So that’s… that’s pretty cool. And what can… what can you tell me about, like, your role at Brainforge, and, like, what’s… what’s going on there, either project-wise or… or day-to-day?
180 00:21:30.490 ⇒ 00:21:32.949 Samuel Roberts: Yeah, so, I kind of mentioned we have…
181 00:21:32.950 ⇒ 00:21:52.579 Samuel Roberts: client work. We don’t have as much AI client work as we do data, and that’s just because we started as a data consultancy, and the AI stuff kind of came about because Utam, who I think you spoke with, is big, big into the, you know, using these tools, and as things are changing, figuring out, building our own, and so some of that had been done, so that platform I mentioned, was kind of
182 00:21:52.580 ⇒ 00:21:55.500 Samuel Roberts: somewhat running, when I joined.
183 00:21:56.610 ⇒ 00:22:02.330 Samuel Roberts: And then I came on because Luton was kind of managing a couple AI engineers,
184 00:22:02.420 ⇒ 00:22:18.289 Samuel Roberts: but is also the, you know, CEO and founder, and has lots of other stuff to be doing, and so I kind of came in to take over that role. A few other clients have come on, and we’re kind of expanding and growing that way, but we’re also focused on the internal tooling. So, it’s kind of this split between the client work, which is…
185 00:22:18.400 ⇒ 00:22:22.559 Samuel Roberts: varies by client, obviously. Different clients are,
186 00:22:22.560 ⇒ 00:22:26.710 Samuel Roberts: more and less interested in different parts of the AI, stack, I guess.
187 00:22:26.710 ⇒ 00:22:47.159 Samuel Roberts: And then internally, we’re just… we’re using these tools, we’re, you know, using Cursor for the most part, you know, we play around with other things, but then we’re also building agents that will, you know, like I said, ingest these meetings and surface things in Slack for you, generate linear tickets and get approvals and things like that. We built a whole tool to do,
188 00:22:47.810 ⇒ 00:23:00.119 Samuel Roberts: case studies. So, the marketing team has been doing, you know, building case studies, and the biggest bottleneck was just scheduling synchronous time, and so we built a little agent, a voice agent, to actually
189 00:23:00.320 ⇒ 00:23:19.019 Samuel Roberts: prompt an engineer, ask questions, do an interview, you know, generate a first pass of that, and get it to marketing. So lots of little… little tools like that that we’re kind of experimenting with. It’s a bit of a push and pull, because the client work is, you know, what pays the bills, and the internal work is what’s helping us accelerate, but that’s kind of my role right now, is a little bit of straddling that.
190 00:23:19.570 ⇒ 00:23:23.059 joaquin: That’s very cool. And what is some of, like, the…
191 00:23:23.430 ⇒ 00:23:35.819 joaquin: client interests in AI, is that kind of, like, a growing trend? Because I think you mentioned that, you know, they’re predominantly data-oriented, but is there any… like, what are some examples of, like, AI solutions that they’re…
192 00:23:35.820 ⇒ 00:23:40.889 Samuel Roberts: Yeah, so I think I mentioned one. One is a chatbot for, a company that,
193 00:23:41.300 ⇒ 00:23:46.249 Samuel Roberts: It’s for their customer service representatives, so we’re not building a customer service
194 00:23:46.360 ⇒ 00:23:56.190 Samuel Roberts: chatbot for users, we’re building a chatbot for the customer service representatives to get a ton of information that they need. So they have documents spread all over the place, they have all this,
195 00:23:56.430 ⇒ 00:24:12.659 Samuel Roberts: information about who gets assigned what, and when a customer calls, and someone needs to go out, who’s in the right region, and so, we’ve built a tool, a chat bot right in Google Chat for them, that can pull that information from, kind of, their docs, or the database of… you know, we had to set up a lot of this database because they were very scattered, but…
196 00:24:13.610 ⇒ 00:24:19.729 Samuel Roberts: That is, that was kind of already ongoing when I joined, so it’s kind of been the longest one. Some other ones were…
197 00:24:19.730 ⇒ 00:24:20.889 joaquin: Did that one go live?
198 00:24:21.490 ⇒ 00:24:25.830 Samuel Roberts: Yeah, it’s live, we’re constantly iterating on it, we’re constantly building it, you know, things are…
199 00:24:26.190 ⇒ 00:24:37.909 Samuel Roberts: they… it’s an interesting project because, again, their data was all over the place, and so we helped centralize a lot of it, but then, are they updating it? Is it fresh enough? Like, there’s a lot of stuff there. Is our… you know, with new models coming out, we’re upgrading that, too.
200 00:24:37.910 ⇒ 00:24:41.629 joaquin: I think, like, a lot of companies face that problem as well, so that’s…
201 00:24:41.630 ⇒ 00:24:56.440 Samuel Roberts: Oh, yeah, cool. Definitely, yeah. Some other ones were… people were, you know, using Claude with a very kind of structured manner. They would copy and paste prompts and take those prompts and paste into other… take those responses and paste into other prompts, and so…
202 00:24:56.440 ⇒ 00:25:07.569 Samuel Roberts: we thought, okay, we can automate that whole thing for you, and just get you that output that you’re trying to get to, and then you can go further with it. So we built a tool like that for an agency.
203 00:25:07.750 ⇒ 00:25:17.070 Samuel Roberts: We were building a platform for another company that, has a lot of clients and has lots of data, and using MCP servers to pull in fresh data and see, and they can get reports and things like that.
204 00:25:17.070 ⇒ 00:25:17.410 joaquin: Hmm.
205 00:25:18.940 ⇒ 00:25:32.689 Samuel Roberts: Yeah, I think really it’s an interesting thing, because a lot of these companies that aren’t necessarily tech companies, or AI, or the, you know, the founders, or the people running them are not technical, but they hear this stuff, they know there’s something there, they’re balancing that
206 00:25:32.730 ⇒ 00:25:41.819 Samuel Roberts: enthusiasm with actually figuring out what problems they have that this can solve, or that we can figure out a good solution for, is an interesting kind of back and forth.
207 00:25:43.880 ⇒ 00:25:55.130 Samuel Roberts: Sorry, I don’t know what we’ll be working on next, you know, clients are kind of always in the pipeline, and there’s some interesting things there. The internal stuff, like I said, is always moving, too. We’re always trying to figure out ways to,
208 00:25:56.370 ⇒ 00:26:09.119 Samuel Roberts: you know, surface things to people that, you know, or simplify processes, or, you know, that case study thing was a good example of just, like, we’re still doing it, it’s just the bottleneck is now a different point.
209 00:26:09.120 ⇒ 00:26:09.730 joaquin: Great.
210 00:26:10.000 ⇒ 00:26:11.730 Samuel Roberts: So, a lot of that stuff.
211 00:26:11.910 ⇒ 00:26:15.140 joaquin: That’s very cool. Yeah, sounds like there’s a lot to work on.
212 00:26:15.140 ⇒ 00:26:16.050 Samuel Roberts: Yes.
213 00:26:16.050 ⇒ 00:26:19.809 joaquin: And, and an exciting time for Brainforge.
214 00:26:19.970 ⇒ 00:26:36.129 Samuel Roberts: Yeah, yeah, definitely. We are… we’re, I joined over the summer, and so it doesn’t… it feels like it’s been way longer at this point, because there’s just so much happening, and we’re growing, headcount and volume of sales and stuff. It’s a really exciting time to be here.
215 00:26:36.510 ⇒ 00:26:55.790 joaquin: Awesome. Last question I’d like to leave off that is just, like, you know, you have these various, different experiences with early-stage startups, I’m sure. A lot of that is, relative to Brainforge, but, like, what do you like about Brainforge? How does it differ from your past experiences?
216 00:26:55.790 ⇒ 00:26:57.900 Samuel Roberts: Yeah, okay.
217 00:26:58.840 ⇒ 00:27:06.840 Samuel Roberts: I… I… I love and hate the, kind of, multiple clients and the internal, like, multiple, kind of, work…
218 00:27:06.990 ⇒ 00:27:14.629 Samuel Roberts: not work streams, because, like, you can have that when you’re building a product, but, you know, context switching is good and bad in some ways, like, it’s nice that it’s a…
219 00:27:14.630 ⇒ 00:27:34.279 Samuel Roberts: okay, I can jump onto something else when I’m stuck somewhere and do something, but it’s also sometimes I’m deep on one thing and I have to get something else done, so there’s a balance there that I’m learning in a way that I didn’t really need to learn before, to this extreme, I guess. Right. The people here are great. You know, we’re all remote, so that can be kind of isolating if you’re not very,
220 00:27:34.280 ⇒ 00:27:41.070 Samuel Roberts: intentional about it, and I think there’s been, and actually, coming from early-stage founder, kind of.
221 00:27:41.070 ⇒ 00:27:55.930 Samuel Roberts: when do you start worrying about culture sort of stuff is probably way earlier than you think, and so I think that’s been thought through a bit more here than, maybe some other companies might have. That’s cool. Speaking of personal experience, I know that to be true, because, you know, that hair care company got to 40 people, and all of a sudden we were like, oh.
222 00:27:55.930 ⇒ 00:28:05.170 Samuel Roberts: crap, what are we doing? We don’t know, you know, all… and the culture forms on its own without you doing anything, so you gotta kind of be on top of that, and I think that’s something that’s intentional here. That’s cool.
223 00:28:05.610 ⇒ 00:28:10.850 Samuel Roberts: Yeah. Trying to think. It’s, the internal work is very exciting, too, it’s almost…
224 00:28:11.000 ⇒ 00:28:11.970 joaquin: It’s, it’s…
225 00:28:11.970 ⇒ 00:28:18.960 Samuel Roberts: It’s neat because we’re using the newest stuff all the time, and playing around with new things, and there’s… you know, it’s less of a fear…
226 00:28:19.080 ⇒ 00:28:34.169 Samuel Roberts: of pushing something on ourselves than it would be to a client. That company goes up in a few years. So there’s definitely considerations there, but it’s a different calculus than client work, and so it’s nice to be able to see a post and say, oh, I’m gonna play with that today and see if it does what we need, and rather than
227 00:28:34.620 ⇒ 00:28:48.859 Samuel Roberts: if it was a client, I’d be like, oh, that’s interesting, but it’s not, you know, quite battle-hardened yet for them. So that ability to just experiment with new things and be on the cutting edge, but also be building, you know, production-ready things is a nice balance.
228 00:28:49.470 ⇒ 00:28:51.360 joaquin: Very cool. Thank you for sharing, Sam.
229 00:28:51.360 ⇒ 00:28:51.940 Samuel Roberts: Totally.
230 00:28:51.940 ⇒ 00:28:53.480 joaquin: That’s all I have for today.
231 00:28:53.480 ⇒ 00:29:04.520 Samuel Roberts: Alright, great, yeah, I think we’re good then. The rest of the process, is, another interview. It’s more role technical, we may have to sort out a little bit how that might, might play out.
232 00:29:04.520 ⇒ 00:29:04.840 joaquin: Yep.
233 00:29:04.840 ⇒ 00:29:07.040 Samuel Roberts: And then after that, I believe, is a panel.
234 00:29:07.160 ⇒ 00:29:10.190 Samuel Roberts: We like to move relatively quick,
235 00:29:10.330 ⇒ 00:29:11.970 Samuel Roberts: So, you know, I should hear back.
236 00:29:12.180 ⇒ 00:29:19.600 Samuel Roberts: Pretty soon, and then I think it’s just scheduling is really the only, holdup there in getting time with people. But, yeah.
237 00:29:19.770 ⇒ 00:29:20.480 joaquin: Sounds good, Sam.
238 00:29:20.480 ⇒ 00:29:21.430 Samuel Roberts: How much for the time?
239 00:29:21.430 ⇒ 00:29:22.909 joaquin: Yup, appreciate it. Have a great day.
240 00:29:23.080 ⇒ 00:29:23.830 Samuel Roberts: Yeah, you as well.
241 00:29:23.850 ⇒ 00:29:24.720 joaquin: Bye-bye.
242 00:29:24.720 ⇒ 00:29:25.250 Samuel Roberts: Bye.