Meeting Title: Brainforge Interview w- Sam Date: 2026-03-12 Meeting participants: Sid Salim, Samuel Roberts
WEBVTT
1 00:03:37.200 ⇒ 00:03:38.230 Samuel Roberts: Hello.
2 00:03:40.640 ⇒ 00:03:41.620 Sid Salim: Hello there.
3 00:03:42.590 ⇒ 00:03:46.979 Samuel Roberts: How are you? Sorry, it was a second late there. Just my computer’s being a little slow today.
4 00:03:47.410 ⇒ 00:03:48.510 Sid Salim: It happens.
5 00:03:48.840 ⇒ 00:03:49.539 Sid Salim: happens, I guess.
6 00:03:49.540 ⇒ 00:03:50.430 Samuel Roberts: Chorus.
7 00:03:51.990 ⇒ 00:03:53.699 Samuel Roberts: How are you doing today?
8 00:03:53.850 ⇒ 00:03:56.290 Sid Salim: I’m doing fine, thanks for asking.
9 00:03:56.870 ⇒ 00:03:57.560 Sid Salim: Love you.
10 00:03:58.470 ⇒ 00:04:00.269 Samuel Roberts: Alright, well,
11 00:04:00.400 ⇒ 00:04:08.820 Samuel Roberts: My name is Sam Roberts, pleased to meet you. I am the AI and automation tech lead here at Brainforge.
12 00:04:11.610 ⇒ 00:04:23.400 Samuel Roberts: just trying to think what you need to know first. Basically, the way this’ll go, I saw your video and application and stuff, so, I have a few questions. This is kind of the first of…
13 00:04:23.940 ⇒ 00:04:31.490 Samuel Roberts: three potential interviews, this one’s not technical, the second one’s more technical, but if we have time, maybe we’ll jump into stuff, we’ll see.
14 00:04:31.840 ⇒ 00:04:36.409 Samuel Roberts: And so, yeah, I think let’s start. If you could just give me a brief introduction, just, you know.
15 00:04:36.750 ⇒ 00:04:40.269 Samuel Roberts: your background, your… who you are. I saw the video, but I would love to hear it, you know.
16 00:04:40.270 ⇒ 00:04:52.740 Sid Salim: live. Sure, no problem. So, my name’s Tid, and I have around 6 years of experience in building, optimizing, and maintaining production AI systems in Python. I’ve worked in healthcare, fintech, and legal data.
17 00:04:52.740 ⇒ 00:05:01.760 Sid Salim: And my main tech stack is Python backend, and over the past few years, my, like, I’ve been working mainly in developing RAG and AI agents using LandGraph.
18 00:05:02.140 ⇒ 00:05:08.260 Sid Salim: But I’ve also implemented, like, traditional AI stuff, like in computer vision, or with the…
19 00:05:08.260 ⇒ 00:05:08.870 Samuel Roberts: Okay.
20 00:05:08.960 ⇒ 00:05:19.559 Sid Salim: with tabular data, you know, XGBoost, and all those sorts of gradient boosting, decision trees and all that. So, that’s a bit about me, like.
21 00:05:20.000 ⇒ 00:05:25.190 Sid Salim: Very brief. I currently work as a senior ML Engineer at Pintech Solutions.
22 00:05:25.450 ⇒ 00:05:32.670 Sid Salim: Before that, I’d worked at Cogenix, and before that, Synergy. Like, these are more B2B, like.
23 00:05:32.840 ⇒ 00:05:52.620 Sid Salim: software companies that work with different clients, and I… like, my current position’s mainly a client-facing position. I interface directly with the clients, like, determine their needs, like, determining what makes it the extra project, and then working on the… working with the data, working on the project, implementing it, delivering it, and…
24 00:05:52.980 ⇒ 00:05:59.679 Sid Salim: optimizing it, you know. Proof of con… sometimes it’s… sometimes, you know, there’s R&D, you have to…
25 00:05:59.880 ⇒ 00:06:10.239 Sid Salim: concept, demonstrate that it’s actually possible, that you can actually use AI to achieve what’s required, and yes, that’s a bit about me?
26 00:06:10.240 ⇒ 00:06:10.840 Samuel Roberts: Great.
27 00:06:11.060 ⇒ 00:06:25.319 Samuel Roberts: Okay, cool, let’s, let’s jump in then. So, client-facing stuff, one thing, you know, I’ve noticed a lot, especially because I don’t come from the consulting world, I come from more of a product startup world, you know, we’re dealing with a lot of non-technical stakeholders.
28 00:06:25.390 ⇒ 00:06:40.999 Samuel Roberts: And so, as you’re definitely aware, the industry is moving so quickly, people hear all kinds of things. How do you explain the limitations of some of this technology to people that aren’t technical, but might have ideas about how they think things should go?
29 00:06:41.750 ⇒ 00:06:46.949 Sid Salim: I honestly think the best way to explain stuff is through charts. Focus on input-out.
30 00:06:48.070 ⇒ 00:06:56.539 Samuel Roberts: Okay. You make charts, and for more complicated things, you may have sub-charts explaining each module. Like, each… you break down each module into simpler…
31 00:06:56.570 ⇒ 00:07:08.889 Sid Salim: Simpler inputs and outputs, and you explain that for this specific input, what sorts of outputs are being generated, what are the outputs that we want, but what outputs you’re getting.
32 00:07:09.350 ⇒ 00:07:20.260 Sid Salim: like, you know, misclassification, you know, an AI model that’s not doing so well, that’s giving us low accuracy results. What those low accuracy results are.
33 00:07:20.520 ⇒ 00:07:25.519 Sid Salim: Why? Why we are experiencing low accuracy? What limitations are there in AI?
34 00:07:25.870 ⇒ 00:07:28.909 Sid Salim: Maybe even talk about what sorts… what…
35 00:07:29.060 ⇒ 00:07:37.720 Sid Salim: the problem with AI models often with the data that we’re dealing with. We may go through regarding what limitations we are facing with our data.
36 00:07:37.960 ⇒ 00:07:41.019 Sid Salim: That’s generally how I’d explain stuff.
37 00:07:41.420 ⇒ 00:07:53.919 Samuel Roberts: Sure. Okay, and then building on that, has there been a time that you can talk about where someone misunderstood what the feature was capable of, or, you know, they had a… can you walk me through a time like that that you can remember?
38 00:07:54.530 ⇒ 00:08:03.659 Sid Salim: Certainly, like, it’s happening at this very moment in time. We’re working on some sort of AI agent that can ex… that can help, like a portfolio manager.
39 00:08:03.800 ⇒ 00:08:09.729 Sid Salim: That is, like, you have a portfolio, and the AI is meant to…
40 00:08:09.850 ⇒ 00:08:14.449 Sid Salim: Improve, like… like, give an idea, like…
41 00:08:14.680 ⇒ 00:08:17.739 Sid Salim: Buy this stock, and sell this stock, and so on and so on.
42 00:08:18.890 ⇒ 00:08:27.489 Sid Salim: And we’re working through this whole project, trying to explain what the big issue is. Like, we are experiencing a bit of a…
43 00:08:27.750 ⇒ 00:08:45.370 Sid Salim: for example, what’s happening in the Strait of Hormuz, and what’s… and its impacts on… on the whole stock market, and how the current data… how our model was trained on old data, and that is not really relevant in this scenario.
44 00:08:45.870 ⇒ 00:08:46.560 Samuel Roberts: Right.
45 00:08:46.560 ⇒ 00:08:57.300 Sid Salim: Because these sorts of shocks was not present in the data that we were modeling it on. And, yes, that’s something that we’re trying to explain.
46 00:08:57.300 ⇒ 00:08:57.840 Samuel Roberts: Great.
47 00:08:58.290 ⇒ 00:09:04.890 Samuel Roberts: Yeah, no, that’s a really good, timely example, yeah. Okay, let’s jump a little bit. Let’s talk about,
48 00:09:05.060 ⇒ 00:09:25.059 Samuel Roberts: So, obviously, I mentioned, like, things are changing so quickly in the industry, and new models are coming out, new tools, frameworks, coding agents and things. Can you tell me about a trend in the industry that you might have been excited about, but for some reason decided it’s not worth adopting right now? Or shouldn’t be adopted, or it’s not ready, or whatever reasons you have?
49 00:09:28.380 ⇒ 00:09:32.840 Sid Salim: There is this awful push towards cloud computing.
50 00:09:33.060 ⇒ 00:09:35.059 Sid Salim: Like, everything’s going…
51 00:09:35.210 ⇒ 00:09:43.519 Sid Salim: to the cloud domain, and while, like, learning AWS, GCP, and all these various cloud technologies, it’s all nice and all, but honestly.
52 00:09:43.740 ⇒ 00:09:51.449 Sid Salim: Me, personally, like, It’s like a car. You’re… instead of purchasing a car, you’re renting it out.
53 00:09:52.560 ⇒ 00:09:53.799 Sid Salim: Know what I mean? Know what I mean?
54 00:09:54.130 ⇒ 00:10:09.320 Sid Salim: While I’m not… while I’m not completely against the idea of working in the cloud, I’d say that wouldn’t it be better if we had our own local… localized, our own server, our own hardware, our own, like, all that?
55 00:10:09.600 ⇒ 00:10:15.590 Sid Salim: And work, and that would… Again, of course, the problem would be in terms of scalability, that.
56 00:10:15.590 ⇒ 00:10:16.330 Samuel Roberts: Sure.
57 00:10:16.330 ⇒ 00:10:25.340 Sid Salim: Sure, of course. That kind of… that’s the… that’s the downside, of course. Again, but what if our product isn’t really that scalable? What if it’s a niche?
58 00:10:26.430 ⇒ 00:10:27.740 Sid Salim: Sure, definitely.
59 00:10:28.050 ⇒ 00:10:35.389 Samuel Roberts: Yeah, interesting. Alright, good. On that same kind of same thought,
60 00:10:35.490 ⇒ 00:10:45.050 Samuel Roberts: So things that are maybe tools you might have seen or played with, like, what… how do you determine if something is production ready to use in a client project, for example?
61 00:10:45.340 ⇒ 00:10:54.650 Samuel Roberts: So, you know, we have internal stuff and external stuff. The internal stuff, we definitely experiment a bit more, but we’re a little more picky about what goes out. So I’m curious where you… how you think about that.
62 00:10:54.650 ⇒ 00:11:13.520 Sid Salim: So, the main things other than accuracy, of course, like, in the production environment, the data that you are training your AI model on is completely different from what the data that your AI model is going to see when it’s in production. And that’s where the whole MLOps thing, MLOps philosophy comes into mind.
63 00:11:13.520 ⇒ 00:11:19.329 Sid Salim: Aside from that, the other metrics that you’d see with your whole,
64 00:11:19.340 ⇒ 00:11:33.689 Sid Salim: whole thing is, basically, there are two parameters, throughput and, latency. Like, there are certain specifications that you’ve already agreed with to the client, that you’re expecting, these many requests per minute.
65 00:11:33.860 ⇒ 00:11:38.980 Sid Salim: And when a client gives an input, it’s… the client expects an output.
66 00:11:39.110 ⇒ 00:11:49.079 Sid Salim: by… by this set amount of time. If it fails… if the client fails to receive an output, well, the client gets bored, and the client’s not going to use our… our application through our support.
67 00:11:49.640 ⇒ 00:11:55.839 Sid Salim: And no matter how accurate our model is, if it doesn’t respond quickly, it’s… No use to us.
68 00:11:56.230 ⇒ 00:12:04.680 Samuel Roberts: Definitely, definitely. Great. So you said, Python, like, Langchain sort of stuff.
69 00:12:04.790 ⇒ 00:12:12.940 Samuel Roberts: So is that… where, I guess, in the AI stack, I guess, have you spent the most time building, overall?
70 00:12:13.650 ⇒ 00:12:15.960 Sid Salim: My main tech stack is Pythonback.
71 00:12:16.670 ⇒ 00:12:21.589 Samuel Roberts: Okay, so Python, Langchain, like, are you… what kind of models do you… are you… are you…
72 00:12:21.590 ⇒ 00:12:22.190 Sid Salim: No.
73 00:12:22.820 ⇒ 00:12:23.720 Samuel Roberts: Talking to.
74 00:12:23.720 ⇒ 00:12:26.190 Sid Salim: iTorch, TensorFlow, that sort of stuff. Okay.
75 00:12:26.190 ⇒ 00:12:26.820 Samuel Roberts: Okay.
76 00:12:26.820 ⇒ 00:12:34.569 Sid Salim: scikit-learn, scikit-learn for… if you’re looking to implement XGBoost, you might use scikit-learn. You could also do that in Spark.
77 00:12:34.840 ⇒ 00:12:36.439 Samuel Roberts: Spark, by Spark?
78 00:12:37.060 ⇒ 00:12:37.440 Samuel Roberts: Sure, sure.
79 00:12:37.620 ⇒ 00:12:55.240 Sid Salim: implement your own MLP, you may implement on using TensorFlow, you may make your own, or TensorFlow or PyTorch. But if you want to use some sort of API, you may want to use Langchain. If you want to use a local LLM on your own thing, I generally use Ullama.
80 00:12:55.450 ⇒ 00:12:57.580 Sid Salim: Importing ulama, and then all that.
81 00:12:58.530 ⇒ 00:12:59.120 Samuel Roberts: Great.
82 00:12:59.910 ⇒ 00:13:03.580 Samuel Roberts: I guess, let’s see…
83 00:13:04.540 ⇒ 00:13:08.280 Samuel Roberts: There’s a whole bunch of questions here, but we’ve covered a lot of it so far.
84 00:13:08.620 ⇒ 00:13:10.150 Samuel Roberts: I guess,
85 00:13:11.030 ⇒ 00:13:19.109 Samuel Roberts: Yeah, I guess let’s switch a little bit. Whatever questions you have for me, I want to make sure we have time, so maybe we can intersperse a little bit and see where it takes us, so…
86 00:13:19.110 ⇒ 00:13:19.530 Sid Salim: Boom!
87 00:13:19.530 ⇒ 00:13:24.559 Samuel Roberts: What can I… what can I tell you about, Brain Forge, the role, any questions you have, yeah?
88 00:13:24.700 ⇒ 00:13:33.259 Sid Salim: I’d like to know what sort of… what sort of projects, would… would I be… suppose I get… I get the job, what sort of projects would I be working?
89 00:13:33.260 ⇒ 00:13:39.690 Samuel Roberts: Yeah, yeah, so, little background, I guess, just to explain. So we have kind of two different engineering teams,
90 00:13:39.850 ⇒ 00:13:54.749 Samuel Roberts: or engineering is split into two, I would say. We have the data side, and so they’re focused a lot on clients with ETLs and data modeling and worrying about a lot of that sort of stuff, so then the AI clients kind of sometimes come out of that, and they’re looking for…
91 00:13:54.810 ⇒ 00:14:05.270 Samuel Roberts: You know, whether that’s, chatbots or, other kinds of automations and things. So we’ve done a few different projects, where we’ve put,
92 00:14:05.780 ⇒ 00:14:17.930 Samuel Roberts: We kind of consolidated a lot of the information that the customer service representatives for a company needs, and so we built a bot that not, you know, end customers are not talking to, but the customer service representatives
93 00:14:17.930 ⇒ 00:14:28.280 Samuel Roberts: rather than going out and finding all these documents all over the place, we set up a RAG pipeline for that. We’ve done some others for people that are getting into using Claude for their work.
94 00:14:28.280 ⇒ 00:14:41.050 Samuel Roberts: or ChatGPT, and had kind of workflows built, but they were just copying and pasting, and so we helped them automate that a little bit so they could get through it faster. We’ve done some other things with,
95 00:14:41.410 ⇒ 00:14:49.959 Samuel Roberts: MCPs and pulling data from ad accounts and analyzing things that way. That was a little bit of a hybrid one, because there was some data stuff there as well.
96 00:14:49.960 ⇒ 00:15:07.090 Samuel Roberts: Then the other side of it is internal tooling, so we’re obviously, using a lot of this tool… a lot of this stuff ourselves. We are, building… we have a platform that ingests all our meetings and the transcripts, so you can search over it and see how the client health is over time. So we work a lot on that stuff as well.
97 00:15:07.210 ⇒ 00:15:21.180 Samuel Roberts: there’s kind of a push and pull between those two, you know, we have the internal, and we think of them as a client, kind of the internal tooling, and then the external clients, and so, yeah, whoever, steps into this role will probably hit the ground running on probably at least one client project, and then some internal stuff, so…
98 00:15:23.090 ⇒ 00:15:23.680 Sid Salim: Nope.
99 00:15:24.220 ⇒ 00:15:43.860 Samuel Roberts: Yeah, and we’re mostly a full-stack kind of place, I’ll just kind of give you some more context there. So we do do some Python, the data side does a lot of Python. We have been experimenting with some Python, we do some… a lot of TypeScript, because we’re building UIs and things as well, so depending on the project, we’ll kind of sometimes stick full-stack TypeScript, if it’s less…
100 00:15:44.080 ⇒ 00:15:48.550 Samuel Roberts: UI-focused, sometimes it’ll go to Python, pipelines. Excuse me.
101 00:15:48.630 ⇒ 00:16:04.309 Samuel Roberts: we’ve done some stuff with N8N, mostly for prototyping and just getting things, you know, tested a little bit. And we built some big things on there, actually, in the past that became a little bit hard to manage, so now we’re moving a lot of that stuff around. But,
102 00:16:05.190 ⇒ 00:16:08.189 Samuel Roberts: Yeah, I think that kind of covers most of that sort of… sort of stuff that way, but yeah.
103 00:16:08.190 ⇒ 00:16:09.760 Sid Salim: Yes, yes. Okay.
104 00:16:09.790 ⇒ 00:16:10.690 Samuel Roberts: What else?
105 00:16:10.880 ⇒ 00:16:15.930 Sid Salim: Let’s see now… That’s, you know…
106 00:16:16.950 ⇒ 00:16:21.830 Sid Salim: That’s actually the biggest thing. Company culture, remote first.
107 00:16:21.830 ⇒ 00:16:30.290 Samuel Roberts: Yeah, this is similar… I was gonna ask you similar questions after this, actually, about, sort of the size of companies you’ve worked at and things, but company culture, we’re fully remote.
108 00:16:31.610 ⇒ 00:16:46.750 Samuel Roberts: we do kind of our check-ins every day, we’re, you know, since we’re a consultancy, we’re meeting with clients, you know, they’re all in Slack with us, but we also have meetings on a weekly basis with them. So, you know, you see people face-to-face, you hear people, you know, we’re on Slack all the time,
109 00:16:46.750 ⇒ 00:17:06.619 Samuel Roberts: you know, we’re growing, so things are kind of… we’re figuring things out, it’s very, you know, we’re a fairly young company overall. But I would say, having done startups before, and having started my own companies and worked at other companies, I think what I really like about, Brainforge is that the culture is being thought about early, because, especially as a fully remote company and a young company.
110 00:17:06.710 ⇒ 00:17:17.190 Samuel Roberts: it could be very easy to not think about that stuff, and end up getting a kind of siloed environment, or a little toxicity, or things like that. And I think, everyone is kind of
111 00:17:17.190 ⇒ 00:17:35.529 Samuel Roberts: worried about that. We want to make sure that we’re getting to know people and, you know, interacting beyond just messages on Slack. So there’s, you know, we do a lot of video chats, we have some stuff where we’re… you know, I’ve chatted with people that I don’t work with just on a kind of 30-minute call sometimes, so, you know, we’re trying to be very intentional about that sort of stuff.
112 00:17:36.380 ⇒ 00:17:41.250 Samuel Roberts: But yeah, that leads me to what I was gonna ask you, is you mentioned some of the companies, I’m curious.
113 00:17:41.420 ⇒ 00:17:46.739 Samuel Roberts: Size, and kind of stage, and maybe how big of teams you’ve been on, and things like that.
114 00:17:46.990 ⇒ 00:17:50.250 Sid Salim: Never pretty small, like, near startups. Okay.
115 00:17:51.730 ⇒ 00:17:56.939 Samuel Roberts: So you’re comfortable in kind of a quick-paced, small team environment? Okay, great, great.
116 00:17:57.450 ⇒ 00:18:05.780 Samuel Roberts: That’s exciting, because that’s definitely what we’ve, what we’re doing. Okay. Other, other thoughts here?
117 00:18:08.050 ⇒ 00:18:15.320 Samuel Roberts: Yeah, I think I got most of my stuff answered. We went through that, that,
118 00:18:16.800 ⇒ 00:18:21.369 Samuel Roberts: Yeah, I think, I think I’m good, so if you don’t have any other questions.
119 00:18:22.010 ⇒ 00:18:24.210 Sid Salim: I’ll reach out to you if I get it.
120 00:18:24.210 ⇒ 00:18:27.560 Samuel Roberts: Okay, great. So yeah, so the next steps,
121 00:18:27.850 ⇒ 00:18:40.400 Samuel Roberts: I’ll bring this back to the team, we’ll go through it. The next stage would be a second, kind of more role-focused, technical kind of interview. After that, there’d be a tech
122 00:18:40.690 ⇒ 00:18:53.169 Samuel Roberts: evaluation, kind of little project, and then a panel kind of interview where you talk about that project with a few of us on the team. And so, we’d like to move relatively quickly, we don’t want to drag these things out,
123 00:18:53.290 ⇒ 00:19:03.289 Samuel Roberts: so you should hear back relatively soon, and then scheduling is just kind of the main bottleneck, is just getting time in front of people. So, yeah, I think that pretty much covers everything.
124 00:19:03.400 ⇒ 00:19:04.500 Samuel Roberts: What do you need here?
125 00:19:04.720 ⇒ 00:19:06.100 Sid Salim: Okay then, well…
126 00:19:06.100 ⇒ 00:19:06.890 Samuel Roberts: Alright.
127 00:19:06.890 ⇒ 00:19:09.030 Sid Salim: Alright?
128 00:19:09.030 ⇒ 00:19:11.040 Samuel Roberts: Thanks so much. Thanks for taking the time.
129 00:19:11.870 ⇒ 00:19:12.430 Sid Salim: Likewise.
130 00:19:12.430 ⇒ 00:19:12.770 Samuel Roberts: Good.
131 00:19:12.770 ⇒ 00:19:13.420 Sid Salim: I met…
132 00:19:13.420 ⇒ 00:19:14.350 Samuel Roberts: Yes, bye.