Meeting Title: Brainforge Interview w- Sam Date: 2026-03-03 Meeting participants: Samuel Roberts, Sowmya
WEBVTT
1 00:00:36.050 ⇒ 00:00:36.910 Samuel Roberts: Hello.
2 00:00:36.910 ⇒ 00:00:39.189 Sowmya: Hi, Sam! How are you doing today?
3 00:00:39.750 ⇒ 00:00:41.090 Samuel Roberts: Good, good. How about yourself?
4 00:00:41.450 ⇒ 00:00:43.720 Sowmya: I’m doing great, thank you for asking.
5 00:00:44.470 ⇒ 00:00:45.530 Samuel Roberts: Of course, of course
6 00:00:45.740 ⇒ 00:00:51.719 Samuel Roberts: Sorry, forgive me, I have a little bit of a tickle in my throat today, so if I’m coughing or something, I apologize.
7 00:00:52.440 ⇒ 00:00:53.590 Sowmya: No problem.
8 00:00:53.730 ⇒ 00:00:56.339 Samuel Roberts: Alright, great. Well, welcome.
9 00:00:57.090 ⇒ 00:01:06.000 Samuel Roberts: Again, so, just brief intro. My name is Sam Roberts. I’m the tech lead for the AI engineering team here at Brainforge.
10 00:01:06.330 ⇒ 00:01:11.959 Samuel Roberts: Yeah, we’ll get into a little bit more about, I guess, maybe that, but I would love for you to introduce yourself, if you don’t mind.
11 00:01:11.960 ⇒ 00:01:23.479 Sowmya: Sure, like, hi. I have, like, close to 5 years of experience building the machine learning systems that actually make it, production and, and get ready.
12 00:01:23.480 ⇒ 00:01:41.910 Sowmya: Use it by retail teams. Most recently at Bank of America, I have been working customer analytics and document intelligence tools, like using large language models. A big part of my work is taking models, like ideas from experimentation to something scalable and stable.
13 00:01:41.910 ⇒ 00:02:01.160 Sowmya: like, working on the data pipelines, building the APIs, and also deploying models, and making sure they have monitored properly. I have also spent time improving the… how we evaluate the AI systems, especially, like, the LLM-based features, so that they are reliable and grounded in the
14 00:02:01.180 ⇒ 00:02:16.060 Sowmya: real data, and that’s all. Previously, I have worked at Infosys. I developed the Transformer-based NLP and also document intelligence platforms using Hugging Face and also Azure OpenAI, deployed them on AWS.
15 00:02:16.500 ⇒ 00:02:26.550 Sowmya: And I am certified in both AWS and Azure, and I’m particularly interested in building these, scalable and real-world AI systems.
16 00:02:26.930 ⇒ 00:02:27.570 Samuel Roberts: Great.
17 00:02:28.670 ⇒ 00:02:29.430 Samuel Roberts: Thanks, thanks.
18 00:02:35.160 ⇒ 00:02:38.050 Sowmya: I think I cannot.
19 00:02:38.050 ⇒ 00:02:43.559 Samuel Roberts: No, I hit my mic. There’s a very sensitive button on it, and I forget that it’s there. It’s a new mic. Thank you.
20 00:02:43.560 ⇒ 00:02:44.190 Sowmya: Okay.
21 00:02:45.650 ⇒ 00:03:00.889 Samuel Roberts: So I would, you talked a lot about the, or you mentioned, I should say, the, some of the stuff you built. I’m curious, could you talk a little bit about the impacts that those had on the business? You know, so, like, the technical side of things is great, but I’d love to hear about the, kind of, ramifications and what have… yeah.
22 00:03:01.690 ⇒ 00:03:26.639 Sowmya: Yeah, sure, like, let me talk about the impact, like, I’m most proud of, like, or, I can say that we are, having the… improve the customer retention by about 15%, which directly translated the, like, multi-million dollar revenue preservation. Like, what made it powerful is, like, it did not just stop at the productions, we integrated the model outputs into CRM,
23 00:03:26.640 ⇒ 00:03:32.530 Sowmya: And call center workflows, so the agents could take the immediate and personalized actions.
24 00:03:32.660 ⇒ 00:03:49.639 Sowmya: That closes the loop between the AI and insights, and business execution. On the operational side, our GPU optimization and auto-scaling also reduced, and infrastructure cost by nearly 20%. Like, that was a big win for the finance team.
25 00:03:49.640 ⇒ 00:04:06.530 Sowmya: And with the hybrid rack pipeline for the complaints team, the, like, audit team saw 22% improvement in document search accuracy, like, which has the review time significantly, cutting the review time. That’s what the impact we have, like.
26 00:04:06.530 ⇒ 00:04:07.370 Sowmya: Nope.
27 00:04:07.610 ⇒ 00:04:09.260 Sowmya: current projects.
28 00:04:10.630 ⇒ 00:04:11.200 Samuel Roberts: Great.
29 00:04:11.490 ⇒ 00:04:15.839 Samuel Roberts: Thank you, thank you. That’s good to hear. I appreciate numbers there, that’s wonderful.
30 00:04:17.160 ⇒ 00:04:30.360 Samuel Roberts: Alright, let’s, tell me about working with non-technical folks. So, obviously, you know, we can communicate about the tech stuff, and then we will in, you know, other interviews will get more technical, but I’m curious about how you think about and operate with
31 00:04:30.540 ⇒ 00:04:38.470 Samuel Roberts: less technical people communicating ideas, getting them to understand, or understanding their problem, whatever it is. Like, how does… how do you think about that, and how does that go?
32 00:04:39.310 ⇒ 00:04:55.589 Sowmya: Yeah, that’s a really interesting one, because it’s something I have learned, like, it has a value a lot over the years, because I think the key is to translate the technical complexity into the business relevance, like, I think, like, for the non-technical stakeholders, or
33 00:04:55.780 ⇒ 00:05:08.599 Sowmya: things, like, I work closely with the marketing and compliance team, who were not technical at all, so instead of talking about, like, I can say in my project, there is a X-Boost hyperparameters or embedding drift.
34 00:05:08.600 ⇒ 00:05:24.810 Sowmya: I would frame it as, here is how this model helps you identify at-risk customers faster, or this pipeline reduces your audit time by 20%, like this, like, I usually start by, like, listening deeply, understanding their pain points or KPIs.
35 00:05:24.810 ⇒ 00:05:43.679 Sowmya: And what success looks like them… for them, so then I map those goals to the technical solution. I also use the visual storytelling, because dashboards, or we can use the sharp explainability plots, or even simple flow diagrams that… to make the AI system’s logic transparent, right?
36 00:05:43.770 ⇒ 00:05:44.830 Sowmya: So…
37 00:05:44.950 ⇒ 00:05:58.670 Sowmya: But also, like, another thing is, that helps is co-creating metrics. For instance, with the compliance team, we defined the retrieval accuracy in business terms. They understood, not just the question and recall, like we usually have.
38 00:05:58.740 ⇒ 00:06:09.370 Sowmya: So, that builds a trust, and also, like, they can understand the… clearly. That’s how, like, I usually, communicate with the non-technical folks.
39 00:06:10.070 ⇒ 00:06:13.760 Sowmya: Great, great. Have you encountered any situations where,
40 00:06:14.040 ⇒ 00:06:31.779 Samuel Roberts: someone non-technical thinks… like, they have a problem, for example, they might not understand exactly what the problem is in a way, or not understand the problem, but they think of a solution already, and you might have to push back or help them reframe it and understanding. Have you had any experience like that?
41 00:06:34.230 ⇒ 00:06:43.199 Sowmya: Yes, I think, like, I do have experience, like, working on this, too. Like, in one situation, I think, like, they are,
42 00:06:43.950 ⇒ 00:07:02.839 Sowmya: they do have, like, every, like, point, like, for what they wanna do, and what solution. And at the time, like, what I did was, like, I tried to start by understanding what actually matters to them, because what problem they are trying to solve, and also, like,
43 00:07:02.850 ⇒ 00:07:20.290 Sowmya: they made, they also care about, or what’s slowing down. Then, I explain, for example, when working with the complex team, compliance teams, I don’t just talk about the embeddings or model tuning, I frame it as in terms of impact, like reducing the audit time.
44 00:07:20.290 ⇒ 00:07:30.390 Sowmya: So, I would also… I can say that, at that time, like, I will try to formulate, and I will involve the feedback loops, then…
45 00:07:30.440 ⇒ 00:07:38.540 Sowmya: I can share the early prototypes, gather input, and adjust based on the real usage, rather than assuming the first version is perfect.
46 00:07:38.940 ⇒ 00:07:40.509 Samuel Roberts: Yeah, like that.
47 00:07:40.940 ⇒ 00:07:43.829 Samuel Roberts: Okay, good, no, that’s… thank you, yes, that’s perfect.
48 00:07:44.900 ⇒ 00:07:53.599 Samuel Roberts: Let’s see, what else can I… I’m curious, we can, you know, I want to leave some time for you to be able to ask questions too, but I’m gonna keep… keep going, if that’s alright.
49 00:07:53.720 ⇒ 00:08:01.640 Samuel Roberts: How are you looking to grow this year? Like, technically or otherwise, like, what are your, kind of, ambitions and goals, I guess, that way?
50 00:08:02.730 ⇒ 00:08:27.700 Sowmya: Yeah, that’s really a good question. So, like, right now, my current focus, I really want to deepen my expertise in building and reliable LLM-powered products, like, especially around evaluation and agent workflows, and also making system more measurable and trustworthy, because I have built a rack system already, but I want to get better at structured evaluation frameworks, and also designing the LLM
51 00:08:27.700 ⇒ 00:08:28.470 Sowmya: features.
52 00:08:28.470 ⇒ 00:08:31.160 Sowmya: That directly tie to the business impact.
53 00:08:31.230 ⇒ 00:08:44.099 Sowmya: Beyond the technical growth, I also want to grow in product thinking, because working closely with the users, understanding their workflows, and also, like, translating it into the smarter AI solutions.
54 00:08:44.260 ⇒ 00:08:50.120 Sowmya: And in long term, I would see someone who can own the AI features, like end-to-end.
55 00:08:50.290 ⇒ 00:08:54.599 Sowmya: Like that. I just wanna make it from idea to do production.
56 00:08:55.600 ⇒ 00:08:57.989 Sowmya: Which makes the meaningful impact, right?
57 00:08:58.430 ⇒ 00:09:01.270 Samuel Roberts: Yeah, okay, good, thank you.
58 00:09:01.600 ⇒ 00:09:04.100 Samuel Roberts: Let me see what else I have here…
59 00:09:04.620 ⇒ 00:09:10.630 Samuel Roberts: Well, let’s, let’s, let’s change. Do you have any questions for me about, the role, or Brainforge, or anything like that?
60 00:09:10.980 ⇒ 00:09:22.039 Sowmya: Yes, like, I do have some couple of questions right now, because, like, what is the biggest challenge right now the team is currently facing with the AI features?
61 00:09:22.040 ⇒ 00:09:22.740 Samuel Roberts: Yeah.
62 00:09:22.850 ⇒ 00:09:29.919 Samuel Roberts: Yeah, that’s a good one. There’s a lot of different stuff. I mean, as I’m sure you’re, you know, aware of, things are changing.
63 00:09:29.920 ⇒ 00:09:30.690 Sowmya: Yeah.
64 00:09:30.690 ⇒ 00:09:36.279 Samuel Roberts: hourly. It’s crazy. And so, what I think is that’s an interesting
65 00:09:36.820 ⇒ 00:09:42.109 Samuel Roberts: problem, I guess. It’s really exciting, there’s new things to use, knowing what
66 00:09:42.430 ⇒ 00:10:00.200 Samuel Roberts: tools to trust at this point is a challenge there. You know, there’s always a shiny new thing, and we’re in an interesting position where we’re doing internal tooling and client work, so there’s a kind of balance there between what we can experiment with for ourselves and what we feel comfortable
67 00:10:00.560 ⇒ 00:10:03.550 Samuel Roberts: putting into production for clients.
68 00:10:03.550 ⇒ 00:10:04.040 Sowmya: Hmm?
69 00:10:04.040 ⇒ 00:10:19.670 Samuel Roberts: I think there’s an element there of the clients also, like, I kind of alluded to this before, where they think they have a solution in mind. They’ve seen things, they’re following things online, and so they know what the… they think they know, maybe, what the cutting edge is, and where they want to be on that, and sometimes there’s a little bit of…
70 00:10:19.780 ⇒ 00:10:39.659 Samuel Roberts: back and forth there, making sure they… we understand exactly what they’re trying to do, rather than what they think they… that’s kind of why I’m curious about that… that side of things. yeah, I mean, there’s, you know, it’s just… there’s a lot of, a lot of moving parts for us right now. You know, we’re a small team doing a lot of things internally and externally, client work and, and,
71 00:10:40.760 ⇒ 00:10:56.230 Samuel Roberts: you know, trying to use the tools that we’re also building the tools. There’s a lot of that kind of, you know, what can we do for ourselves internally as a company? What can we speed up? How do we translate that, maybe, to client work?
72 00:10:56.790 ⇒ 00:11:10.770 Samuel Roberts: you know, in terms of agents, you know, there’s all kinds of… evaluations is a big thing for us, I’m glad you mentioned that, because I think there’s a lot of places we could be doing much better evaluations, and I think that’s, you know, we have some, but every time we kind of build new things.
73 00:11:10.770 ⇒ 00:11:18.090 Samuel Roberts: that’s a piece that sometimes might not get the full focus, we’ll say. You know, getting something out is exciting, but then kind of…
74 00:11:18.730 ⇒ 00:11:25.150 Samuel Roberts: strengthening it and solidifying it sometimes, doesn’t always happen as quickly as it could.
75 00:11:25.690 ⇒ 00:11:30.460 Samuel Roberts: Yeah, I hope that… I don’t know if that answers the question well enough. There’s a lot of, you know, I could go into more detail, but I’m trying to… I don’t want to get too…
76 00:11:30.460 ⇒ 00:11:31.420 Sowmya: Yeah, yeah, go ahead.
77 00:11:31.420 ⇒ 00:11:35.240 Samuel Roberts: Technically, at least, but… Okay, great, great. Yeah.
78 00:11:35.240 ⇒ 00:11:41.999 Sowmya: Yeah, and like, if I were joined, like, what do you want me to focus on first? Like, is there any…
79 00:11:42.680 ⇒ 00:11:45.150 Sowmya: thing that I need to focus, yeah.
80 00:11:47.980 ⇒ 00:11:59.419 Samuel Roberts: we’re all kind of, I don’t want to say generalists when it comes to the AI stuff, but we don’t necessarily have, like, a specific, like, this person only does this kind of thing, or this kind of agent. You know, that may change as we grow. Like I said, we’re kind of a small team.
81 00:11:59.830 ⇒ 00:12:05.210 Samuel Roberts: kind of the nature of it right now. I mean, your mention of evaluations is, again.
82 00:12:05.580 ⇒ 00:12:06.859 Samuel Roberts: Something I’m… I’m…
83 00:12:06.960 ⇒ 00:12:17.590 Samuel Roberts: I’m excited to hear a little more about as the process may go on, because I think we… we have some… we have a little… we’re a little stronger, maybe, with some of the implementation stuff, and less of the,
84 00:12:17.730 ⇒ 00:12:31.899 Samuel Roberts: testing and evaluation stuff, and so I think there’s a lot that we can do there. You know, we do some, but again, it’s, you know, we only have so much time and things like that, so… but especially for client work, there’s a lot of stuff where I think we have some kind of, you know,
85 00:12:32.360 ⇒ 00:12:35.789 Samuel Roberts: Initial evaluations, and we have some ongoing things that could be
86 00:12:36.120 ⇒ 00:12:42.990 Samuel Roberts: strengthened and solidified and, really hardened in a way that I think would make them
87 00:12:43.370 ⇒ 00:12:46.730 Samuel Roberts: Better systems, and we could improve the systems more.
88 00:12:47.710 ⇒ 00:12:48.520 Samuel Roberts: Consistently.
89 00:12:48.520 ⇒ 00:12:50.700 Sowmya: Okay, that’s it. That is exactly…
90 00:12:50.700 ⇒ 00:13:03.960 Samuel Roberts: Yeah, yeah, yeah, I don’t… again, this is just me thinking about it right now. I don’t know if that’s exactly what it would be. And again, there’s also the interesting thing about, you know, this is sort of… I come from a startup world, a kind of a more product world.
91 00:13:03.960 ⇒ 00:13:16.939 Samuel Roberts: This is more client-based, so we’re doing some product work, but it’s… there’s client projects, and so it’s a little bit… you bounce around a lot, so there might be different things depending on, you know, what the clients need, and so we have a number of, kind of, different sort of projects. It’s not…
92 00:13:16.950 ⇒ 00:13:19.560 Samuel Roberts: the same thing every time yet.
93 00:13:19.970 ⇒ 00:13:22.849 Samuel Roberts: So that might, you know… you’ll be bouncing around a lot, I imagine.
94 00:13:23.190 ⇒ 00:13:24.170 Samuel Roberts: Which are done very far.
95 00:13:24.170 ⇒ 00:13:28.899 Sowmya: Got you, got you. Like, that’s exciting to hear, like,
96 00:13:29.080 ⇒ 00:13:31.720 Sowmya: Like, something we can confidently ship.
97 00:13:32.030 ⇒ 00:13:32.710 Samuel Roberts: Yeah.
98 00:13:33.690 ⇒ 00:13:36.660 Samuel Roberts: Definitely, definitely. Okay, yeah, what else, what else can I…
99 00:13:36.660 ⇒ 00:13:44.590 Sowmya: That’s all, like, what the… finally, like, what would be the next steps in the process look like, yeah.
100 00:13:44.590 ⇒ 00:13:54.850 Samuel Roberts: the usual question we ask? Of course, of course, yeah, no, I was definitely gonna mention that if you hadn’t asked, but yeah, so this is, so I believe you submitted an application with a loom, and that kind of…
101 00:13:54.850 ⇒ 00:14:04.710 Samuel Roberts: you know, got reviewed and got you to this point. And so this is kind of the first interview. It’s usually more of a chat like this. It’s less, not quite as technical as maybe the
102 00:14:04.710 ⇒ 00:14:14.710 Samuel Roberts: next one may be. So then there’ll be a second round, with another person, and I, I believe, there will be a little bit of a tech,
103 00:14:14.740 ⇒ 00:14:17.129 Samuel Roberts: Evaluation, either before
104 00:14:17.210 ⇒ 00:14:21.510 Samuel Roberts: Probably after that one? Actually, I don’t remember exactly where the order is here, but you would get something.
105 00:14:21.510 ⇒ 00:14:22.310 Sowmya: Exactly.
106 00:14:22.310 ⇒ 00:14:26.189 Samuel Roberts: Yeah, exactly. And then the final round is more of a,
107 00:14:26.380 ⇒ 00:14:44.040 Samuel Roberts: like a panel interview, so I think I’d be there, and I think the second interviewer, and then someone else, so we would have a little more of a chat, and probably talk through the technical, evaluation a bit more. And then from there, you know, if you make it all the way through each of those kind of checkpoints, yeah, that would be… that would be it. So.
108 00:14:44.040 ⇒ 00:14:44.580 Sowmya: Okay.
109 00:14:44.580 ⇒ 00:14:44.980 Samuel Roberts: But…
110 00:14:44.980 ⇒ 00:14:46.290 Sowmya: That sounds to me.
111 00:14:46.290 ⇒ 00:14:52.389 Samuel Roberts: We try to move relatively quickly. I think, you know, scheduling these kinds of face-to-face things is the hardest part, just because.
112 00:14:52.390 ⇒ 00:14:53.120 Sowmya: People’s schedules.
113 00:14:53.120 ⇒ 00:14:55.260 Samuel Roberts: but, you know, we don’t like to drag things out, so…
114 00:14:55.340 ⇒ 00:15:12.139 Sowmya: Yeah, got you. Like, it was very interesting, like, to hear about your projects and, current ongoing challenges. Like, I think, I really enjoyed our conversation today, and, like, if you have more questions, I can happy to answer right now.
115 00:15:12.590 ⇒ 00:15:22.149 Samuel Roberts: Okay, yeah, I… you know, I would love to hear a little bit about, working on a team, working on the same kind of, you know, project, how, how…
116 00:15:22.150 ⇒ 00:15:37.370 Samuel Roberts: you think about, maybe, you know, your tasks and other people’s tasks, and when they inter… interact, I guess, when some of your code affects other code, and vice versa. What does that, how does that look? When does that go well? When does that go poorly? You know, what have you kind of taken from that, if that makes sense?
117 00:15:38.160 ⇒ 00:15:57.439 Sowmya: Okay, got you, like, normally, like, whenever working in the same project, I try to think, in terms of ownership and clear interfaces. Like, if my code affects someone else, we define the contracts early, like, API schemas, data formats, or expected outputs, so we are not guessing later, right?
118 00:15:57.440 ⇒ 00:15:57.950 Samuel Roberts: Perfect, yeah.
119 00:15:57.950 ⇒ 00:16:14.920 Sowmya: It goes well when communication is proactive. For example, if I am changing something that could impact the downstream system, I will flag it early and align before merging. It goes poorly when assumptions are not discussed, like, especially around edge cases or performance expectations.
120 00:16:14.940 ⇒ 00:16:30.100 Sowmya: What I have learned is, like, is that small things at least save a lot of debugging later. I try to document the decision, keep PRs clear, and also test the integration points carefully before reviewing.
121 00:16:30.640 ⇒ 00:16:34.439 Samuel Roberts: Okay, great, great. Let’s see what else,
122 00:16:37.020 ⇒ 00:16:41.440 Samuel Roberts: That kind of covers most of what I was looking to ask.
123 00:16:41.870 ⇒ 00:16:43.040 Samuel Roberts: I think…
124 00:16:43.640 ⇒ 00:17:00.040 Samuel Roberts: Yeah, we’ve covered a lot about that. I actually, I mean, so again, I didn’t want to get too technical on this one, but we have a little bit of time, so I’m curious, on what tools have you, you know, used, what kind of, you know, tooling for agents and, evaluations and things have you… are you most familiar with, I guess?
125 00:17:00.430 ⇒ 00:17:15.219 Sowmya: Yeah, like, the tools, like, what I have used, like, most recently for my projects, like, I’m comfortable with Python-based ML and LLM systems, like, especially building rack pipeline, evaluation workflows, and production APIs using FastAPI.
126 00:17:15.220 ⇒ 00:17:23.449 Sowmya: And I have worked a lot with Azure OpenAI and Hugging Face Transformers and deployed systems on Kubernetes in AWS environments.
127 00:17:23.530 ⇒ 00:17:39.020 Sowmya: My strength is really in designing end-to-end LLM features, you know, like, currently we have some, frameworks, orchestration frameworks like Langtrain, Langraph, and those all things. Like, I have worked on these, kind of orchestration tools.
128 00:17:39.020 ⇒ 00:17:41.839 Samuel Roberts: Okay, great. Yeah, we’ve used a little bit of,
129 00:17:41.840 ⇒ 00:18:03.650 Samuel Roberts: BlindChain, a little, you know, we do Python, you know, we have a whole kind of data side of the company, there’s a lot of Python. We use a lot of TypeScript right now, because we’re building kind of, you know, web UI features as well, so it’s nice to have that kind of consistency, but, you know, we don’t necessarily think there’s one right way to do, you know, there’s not necessarily the one right tool for every job kind of thing. There’s different tools, and so,
130 00:18:03.650 ⇒ 00:18:07.780 Samuel Roberts: You know, we are open to experimentation that way as well, especially internally.
131 00:18:07.780 ⇒ 00:18:21.530 Samuel Roberts: Where we’ll try out new things, but, again, client work is a little different, you know, it’s just gotta be… but it’s even… even with that, we’re still trying out new things, we’re trying to build new features for people that are not, not already solved problems, you know, new, new things.
132 00:18:21.530 ⇒ 00:18:39.469 Sowmya: Yeah, got you. Like, I do have, like, worked with the TypeScript and those, like, other, files as well, like, that makes a lot of sense. Like, I like that approach, like, choosing tools based on the problem instead of just forcing one task everywhere. Exactly, exactly. Yeah.
133 00:18:39.620 ⇒ 00:18:41.090 Samuel Roberts: Okay,
134 00:18:41.780 ⇒ 00:18:46.299 Samuel Roberts: That covers most of what I think I had, unless you have any other questions, I think we can…
135 00:18:47.300 ⇒ 00:18:48.109 Samuel Roberts: I’m happy to answer.
136 00:18:48.110 ⇒ 00:18:48.630 Sowmya: No.
137 00:18:48.630 ⇒ 00:18:49.789 Samuel Roberts: have, but yeah.
138 00:18:49.790 ⇒ 00:19:06.349 Sowmya: Yeah, like, I think we covered almost everything, but like, finally, like, I just want to know, as you told the focus, like, what would you think, like, I should have these qualities so that you will make sure that you have either best person?
139 00:19:06.880 ⇒ 00:19:16.000 Samuel Roberts: Yeah, like, qualities we look for, kind of, in… in… in general, I guess. I kind of mentioned we’re,
140 00:19:16.690 ⇒ 00:19:21.450 Samuel Roberts: Yeah, generalist may not be quite the right word, because we’re obviously focused on it, but we’re not, you know…
141 00:19:21.790 ⇒ 00:19:32.849 Samuel Roberts: oh, I need this done, it goes to that person, because that’s all they do. So I think someone that can kind of… you know, you sort of mentioned end-to-end features, you know, we want someone that can kind of run with it, in…
142 00:19:33.300 ⇒ 00:19:46.980 Samuel Roberts: tackle different parts of the problem along the way. It’s not like, oh, okay, I did my… you know, I wrote the prompt, and I got the agent running, and now I pass it off to someone else. Like, someone that can kind of, you know, shepherd the thing all the way is very important for us.
143 00:19:47.350 ⇒ 00:19:48.150 Samuel Roberts: I think…
144 00:19:48.150 ⇒ 00:19:48.860 Sowmya: Oops.
145 00:19:49.550 ⇒ 00:19:53.760 Samuel Roberts: Flexibility in terms of tooling, which we just kind of talked about, is a big one, because we’re, you know, not only are we
146 00:19:54.550 ⇒ 00:20:03.460 Samuel Roberts: Looking for the right tool for the job, but we’re also experimenting with new things, so ability to learn and not be so…
147 00:20:03.830 ⇒ 00:20:19.550 Samuel Roberts: what, I guess, rigid, in terms of, like, no, no, no, I need this, I need this, you know, we’re constantly reworking our internal processes, we’re constantly reworking, the tooling we’re using, we’re building our own tooling, we’re doing all kinds of things, and so ability to be, flexible that way is really important.
148 00:20:19.550 ⇒ 00:20:28.910 Samuel Roberts: I think in… along with that, just an eagerness to learn and interest in the current status of the industry, and how things are changing, you know, this is not a…
149 00:20:28.910 ⇒ 00:20:29.730 Sowmya: Yeah.
150 00:20:29.730 ⇒ 00:20:39.840 Samuel Roberts: a place where you can be like, oh yeah, I know my, you know, my programming language got written, and I know all the… and, like, LLMs have changed everything in that way, like, there’s a lot of.
151 00:20:39.840 ⇒ 00:20:40.690 Sowmya: That’s right.
152 00:20:40.690 ⇒ 00:20:43.719 Samuel Roberts: quickly, yeah. So I think just that kind of,
153 00:20:43.920 ⇒ 00:20:52.910 Samuel Roberts: yeah, flexibility and eagerness to learn, is kind of, you know, they’re a little more abstract, maybe, again, I don’t want to get too technical in terms of, like, that’ll be down the path.
154 00:20:52.910 ⇒ 00:21:03.400 Sowmya: Yeah, I know, like, you know, like, we can see the… recently we have, like, many more updates coming in the SNDKI and everything, like, we are seeing open floor bots, nanobots, and, like.
155 00:21:03.400 ⇒ 00:21:16.709 Samuel Roberts: Exactly, yeah. Yes, exactly. And so, following along with that stuff is great. You know, we have Slack channels where people are just posting interesting articles and, you know, engaging on that, like, kind of… it’s related to work, but it’s not quite the work, so I think, you know, being.
156 00:21:16.710 ⇒ 00:21:24.390 Sowmya: Yeah, it is interesting, because it’s related to our part only, because it looks interesting when we come across those old things, right?
157 00:21:24.390 ⇒ 00:21:28.239 Samuel Roberts: Exactly, exactly. So, you know, I think being,
158 00:21:28.590 ⇒ 00:21:39.360 Samuel Roberts: get excited about new things is good there. And then, yeah, obviously technical ability is kind of the… you know, I’m not quite touching on it, because I like to keep this a little bit lighter, but, you know,
159 00:21:39.400 ⇒ 00:21:49.440 Samuel Roberts: And I mean technical ability in that you can understand things, you can figure things out, you may not be, you know, you don’t have everything memorized, necessarily. You don’t need that, like, you need to be able to…
160 00:21:49.440 ⇒ 00:21:50.130 Sowmya: Awesome.
161 00:21:50.130 ⇒ 00:21:54.039 Samuel Roberts: Especially now with coding agents, which we haven’t necessarily talked about.
162 00:21:54.380 ⇒ 00:21:57.229 Samuel Roberts: You know, the ability to guide a coding agent.
163 00:21:57.230 ⇒ 00:21:57.710 Sowmya: account.
164 00:21:57.710 ⇒ 00:22:05.850 Samuel Roberts: might… is a different skill, maybe, than writing code, and so I think, there’s a… there’s a relation there, where, you know.
165 00:22:05.850 ⇒ 00:22:18.920 Samuel Roberts: But the kind of taking a step back, or sitting higher up and thinking, okay, I know what needs to get done, I have ideas, I don’t need to implement it necessarily, but I need to know it works, I need to know it’s the right way. And so I think there’s a little bit of a different
166 00:22:19.110 ⇒ 00:22:31.120 Samuel Roberts: mental model, maybe, for how we’re using these tools, but what at the base of that is still, like, a technical expertise or understanding, you know? Like, we don’t write in assembly language, we don’t write, you know, low-level things.
167 00:22:31.120 ⇒ 00:22:36.660 Sowmya: Thing is changed. Exactly, so it’s just… but, like, the understanding, the, you know…
168 00:22:36.660 ⇒ 00:22:49.610 Samuel Roberts: I don’t know the details of assembly language, but I still know, basically, how things work. And so, like, understanding different levels, and being able to use that, and debug things, especially, that’s gonna be a big one. It has been a big one, I should say, but,
169 00:22:49.610 ⇒ 00:23:00.759 Samuel Roberts: That’s right, understanding the flow is very much important, and knowing everything pinpoint. Exactly, exactly. Like, we’re almost designing these black boxes sometimes, and when they break.
170 00:23:00.960 ⇒ 00:23:06.250 Samuel Roberts: We have to pull them apart, and we didn’t write them, you know, it’s like someone else’s code, but,
171 00:23:06.430 ⇒ 00:23:17.569 Samuel Roberts: I think there’s a lot there that is, still very valuable from the, like, understanding, the technical understanding side of things, so that’s kind of another thing we look for a lot, so, that’s a big, important one. Yeah.
172 00:23:18.390 ⇒ 00:23:22.249 Sowmya: God, you… Thank you so much, like, for explaining everything.
173 00:23:22.250 ⇒ 00:23:31.590 Samuel Roberts: Yeah, yeah, no, I’m glad, this is, this is nice, thank you for asking interesting questions. So, yeah, if you have any other questions, I’m happy to answer more, but we’re getting a little close, but yeah.
174 00:23:31.590 ⇒ 00:23:35.069 Sowmya: I don’t have any more questions right now. Okay, great.
175 00:23:35.070 ⇒ 00:23:37.379 Samuel Roberts: And I think… I think we’re all set.
176 00:23:37.380 ⇒ 00:23:38.430 Sowmya: I think…
177 00:23:38.870 ⇒ 00:23:39.960 Sowmya: your time?
178 00:23:40.260 ⇒ 00:23:53.429 Samuel Roberts: Of course, yeah, my pleasure, my pleasure. I believe someone will be in touch one way or another, so either schedule the next one or, you know, whatever happens, but I will synthesize my thoughts and get you moving along, so hopefully nothing takes too long.
179 00:23:54.010 ⇒ 00:23:55.240 Samuel Roberts: Yeah, I think that’s all.
180 00:23:55.240 ⇒ 00:23:55.950 Sowmya: That’s all!
181 00:23:56.220 ⇒ 00:24:01.519 Samuel Roberts: Let me know if you have any other questions before the next interview or anything, and yeah, best of luck in everything.
182 00:24:01.960 ⇒ 00:24:04.380 Sowmya: Thank you, bye, you have a nice day.
183 00:24:04.380 ⇒ 00:24:05.150 Samuel Roberts: You as well.