Meeting Title: Brainforge AI Tech Lead Interview Date: 2026-03-16 Meeting participants: Dan Hartley, Samuel Roberts
WEBVTT
1 00:01:59.120 ⇒ 00:02:00.340 Dan Hartley: Hey, Samuel.
2 00:02:05.320 ⇒ 00:02:06.120 Samuel Roberts: Hi there.
3 00:02:06.810 ⇒ 00:02:07.330 Dan Hartley: Hey!
4 00:02:07.330 ⇒ 00:02:09.070 Samuel Roberts: Let me get my… there we go.
5 00:02:12.040 ⇒ 00:02:16.290 Samuel Roberts: There we go. Alright, sorry about that. Sometimes I have some audio issues on my setup.
6 00:02:16.860 ⇒ 00:02:18.330 Dan Hartley: Sorry.
7 00:02:19.730 ⇒ 00:02:20.680 Samuel Roberts: How are you today?
8 00:02:20.830 ⇒ 00:02:22.429 Dan Hartley: I’m doing great, how are you?
9 00:02:23.500 ⇒ 00:02:26.059 Samuel Roberts: Good, good, yeah. Thanks for taking the time.
10 00:02:27.570 ⇒ 00:02:30.949 Samuel Roberts: Let me just get all my stuff organized here. So, okay,
11 00:02:32.110 ⇒ 00:02:38.209 Samuel Roberts: So my name is Sam Roberts, nice to meet you. I’m the AI tech lead here at Brainforge.
12 00:02:39.180 ⇒ 00:02:54.210 Samuel Roberts: Yeah, so I think the way this… this is the first interview, it’s more just kind of chat, the next one’s more technical, and then there’s a panel interview after that. So I think for this one, I’ve got some questions here. I want to make…
13 00:02:54.210 ⇒ 00:03:01.330 Samuel Roberts: Make sure that there’s time for you to ask whatever questions you have, so probably, like, halfway, we can kind of switch and just kind of back and forth a little bit. But,
14 00:03:01.560 ⇒ 00:03:09.770 Samuel Roberts: let’s, I think I would just start with a, little intro from you, you know, your background briefly, and then, jump into questions. So, yeah.
15 00:03:09.770 ⇒ 00:03:28.469 Dan Hartley: Definitely, definitely, definitely. So, really nice to meet you, Samuel. about me, I’ve started… I’ve been a machine learning engineer for the past 7 or 8 years. In my most recent experience, I’m currently working as a team lead as well. I am leading a team of 6 engineers. Ideally, I think it’s 6 to 7 to 8. Yeah, depends. So…
16 00:03:28.470 ⇒ 00:03:39.449 Dan Hartley: And yeah, I mean, like, the experience has been great. I’ve had the privilege of working in almost all domains of AI. Started off as a software engineer, and then I transitioned towards the machine learning space.
17 00:03:39.520 ⇒ 00:03:58.040 Dan Hartley: And till then, the idea has been amazing. So, yes, I’m really keen on driving and working in startup environments, because obviously, you know, you gotta wear multiple hats at the same time, and that is what excites me a lot, and then that is what motivates me a lot, to just, like, wake up every day and just, like, go back to work and work and work. So, yes.
18 00:03:59.450 ⇒ 00:04:00.240 Samuel Roberts: Cool.
19 00:04:00.240 ⇒ 00:04:23.120 Samuel Roberts: Great, yeah, that’s, that’s awesome. I… very similar boat, where I started, not in this area. I actually… my background is in mechanical engineering to start, and then I got into software, got into web tech, and now it’s much more AI in focus. Cool. Alright, let’s, let’s just jump in. There’s a whole series of questions, I’m just not sure where I want to start best, but…
20 00:04:23.120 ⇒ 00:04:24.939 Samuel Roberts: I guess let’s talk,
21 00:04:25.390 ⇒ 00:04:32.689 Samuel Roberts: Tell me about an LLM-based feature, that you’ve shipped, what the problem solved, and then I have a few more follow-ups after that, so… yes.
22 00:04:32.690 ⇒ 00:04:49.469 Dan Hartley: Definitely, definitely. So, I’ve worked on a lot of LMs, based on, like, a lot of RAGs. So, what I wanted was, I documented, I worked on a document to structure data, so you get to… there was, like, a lot of PDFs, a lot of documents, and then…
23 00:04:49.710 ⇒ 00:05:05.649 Dan Hartley: it was basically kind of a rag, so what it did was, basically, I had to just, like, chunk them, embed them, and then just, like, the… there was an entire LLM layer for just, like… so there was document indecision to the intra… then there was the extraction service, and then came then the part where we had to just, like, ship it out, say yes.
24 00:05:07.180 ⇒ 00:05:07.750 Samuel Roberts: Cool.
25 00:05:08.000 ⇒ 00:05:14.290 Samuel Roberts: Okay, so what… and what was it, like, specifically? Like, was there… the problem that it was solving? Like, was it…
26 00:05:15.150 ⇒ 00:05:15.970 Dan Hartley: Yes, whispering.
27 00:05:15.970 ⇒ 00:05:18.100 Samuel Roberts: data for whatever… yeah, go ahead.
28 00:05:18.100 ⇒ 00:05:32.589 Dan Hartley: Yeah, so it was for a client, and then, basically, they want some… a lot of… they want some information to be extracted out via chatbot, so what they wanted was a lot of virtual data, and then it had to be just, like, frictioned to a virtual database, and then they were just, like, there was some, like, Q&A automation for all of that.
29 00:05:33.000 ⇒ 00:05:40.330 Samuel Roberts: Cool, okay. So, with that, then, specifically, where have you spent the most time building within the kind of, like, AI stack? Like, was it just…
30 00:05:40.450 ⇒ 00:05:42.849 Samuel Roberts: end-to-end? Was it pieces? Like, where do you feel…
31 00:05:42.850 ⇒ 00:05:43.230 Dan Hartley: By the way.
32 00:05:43.230 ⇒ 00:05:46.059 Samuel Roberts: Versus… versus, like, more experimenting with, I guess, you know?
33 00:05:46.060 ⇒ 00:05:46.640 Dan Hartley: Huh.
34 00:05:46.870 ⇒ 00:05:55.799 Dan Hartley: So it was, it was more end-to-end-to-end, right? So I worked on this project, it was not an end-to-end, so I worked on the LM orchestration, crafted the entire prompt.
35 00:05:55.800 ⇒ 00:06:17.650 Dan Hartley: In terms of data, how do I just, like, chunk it, and how do I just, like, embed it, and then just, like, store it out? I also worked on the part where there was… because there was just, like, different and different kind of just, like, documents, right? It was not standardized, right? So just, like, standardizing them to a current format, creating retrievals and context pipelines, and then, yeah, productions.
36 00:06:19.600 ⇒ 00:06:23.230 Samuel Roberts: Cool, okay. So let’s, let’s talk about, like,
37 00:06:23.620 ⇒ 00:06:29.289 Samuel Roberts: how it goes when you’re communicating with stakeholders that aren’t technical. So, obviously, like.
38 00:06:29.600 ⇒ 00:06:37.320 Samuel Roberts: there’s non-technical people, we’ve always had to deal with that as software engineers, but now I feel like with AI, there’s a lot more news out there, and people think
39 00:06:37.430 ⇒ 00:06:43.340 Samuel Roberts: And have all kinds of ideas. So how do you go about kind of explaining the limitations of some of the technology to people that might not be familiar?
40 00:06:43.340 ⇒ 00:06:56.030 Dan Hartley: I think that’s a great question. That is a great question. So, and obviously, I believe in explainable AI, right? So, if there is a prediction, or if there is an answer, and then… and obviously, if it’s not explainable, I don’t think there’s a good use to it.
41 00:06:56.030 ⇒ 00:07:03.680 Dan Hartley: So, I… while communicating to non-technical stakeholders, I try to make sure that I am communicating to them in terms of the
42 00:07:03.680 ⇒ 00:07:04.580 Dan Hartley: business
43 00:07:04.580 ⇒ 00:07:15.179 Dan Hartley: In terms of how it’s going to be effective towards the business, right? So what I try to do is I focus on setting expectations through examples, right? So where… so instances where the model
44 00:07:15.180 ⇒ 00:07:26.059 Dan Hartley: performs really well, instances where the model might hallucinate or fail, and then calculating the cost of it as well, and asking them, what’s the cost of a specific failure? Because that’s important, right? And then…
45 00:07:26.060 ⇒ 00:07:40.629 Dan Hartley: the risks, what, what are the types that are appropriate, what is risky, and then how I go by it. If it’s really a lot, like, a complete non-technical client, then what I try to do is just, like, or stakeholder, then I try to do is just, like, create slides.
46 00:07:40.630 ⇒ 00:07:58.290 Dan Hartley: So, create some PowerPoint slides, add in less… less… add in more images so that it’s more understandable, and then, draw some dashboards around, just, like, pull them out, just, like, sort of then versus now, like, versus what’s next. That’s one of the ideologies that I really prefer, and then, yeah, that’s how I go by it.
47 00:07:59.620 ⇒ 00:08:09.120 Samuel Roberts: Great. Okay, so then building on that, is there an example you can talk about where someone, a user, a client, stakeholder, misunderstood what the feature could do?
48 00:08:09.390 ⇒ 00:08:11.170 Samuel Roberts: And how… how that went.
49 00:08:11.580 ⇒ 00:08:19.619 Dan Hartley: Definitely. So, I have a great story related to it, so if you want, I can just, like, tell you the entire story. I’m not sure if it’s.
50 00:08:19.620 ⇒ 00:08:20.480 Samuel Roberts: Oh, yeah, yeah.
51 00:08:20.820 ⇒ 00:08:37.920 Dan Hartley: Yeah, so it was an end-to-end product ID agent, right? So, for example, there was a client that hopped in, and he was like, hey, I’ve built a trimmer. What it does is it straps onto your neck, and then it has a rail, and then there is a trimmer that adjusts itself onto it. And then,
52 00:08:37.919 ⇒ 00:08:43.129 Dan Hartley: the guard, I programmed it in a way that it can adjust itself, its own, by its own.
53 00:08:43.200 ⇒ 00:08:46.849 Dan Hartley: I was like, okay, fine. So, guess what he wanted me to do next?
54 00:08:49.700 ⇒ 00:08:50.269 Samuel Roberts: I don’t know.
55 00:08:50.270 ⇒ 00:08:58.320 Dan Hartley: Alright, so what you wanted to say, hey, I want to build a system that could… I need AI to cut people’s hair, automatically. I was like, okay.
56 00:08:58.320 ⇒ 00:08:58.700 Samuel Roberts: Okay.
57 00:08:58.700 ⇒ 00:09:23.189 Dan Hartley: That’s interesting, that’s really interesting. So, how are we gonna go by it? So I was like, okay, so let’s start off by building a mobile application and an interface where we can just, like, users can communicate, they can hop in, and then we’re gonna have a feature inside of it, what you can do is just, like, we can send those, send over those values, and then it is going to have a camera that could just, like, draw it, it can use some computer vision right onto it, and then it’s going to draw some gourdles around the, area of the interest, and then it’s going to just, like, go ahead.
58 00:09:23.330 ⇒ 00:09:27.059 Dan Hartley: Okay. And misunderstood it to be that…
59 00:09:27.380 ⇒ 00:09:30.779 Dan Hartley: All of this is just gonna come at a very low cost.
60 00:09:30.780 ⇒ 00:09:55.630 Dan Hartley: And it’s gonna be… it’s gonna be there, right? So, and that… that was a challenge. So what I did was, I came back, drawing, and she’ll draw the entire feasibility around it, did some research, printed out some models which I can integrate, crafted the entire architectural diagram, or how I do want it, just like, what’s the flow of… what is the entire flow gonna be? And then, when it went back to the client, I said, hey, this is the entire crafted flow.
61 00:09:55.630 ⇒ 00:10:06.490 Dan Hartley: here is the entire user journey on how it’s gonna happen, but what’s the drawback is that for each and every trimmer, you would need an integrated GPU inside of your,
62 00:10:06.490 ⇒ 00:10:19.080 Dan Hartley: either we go in by the cloud, but if you want it, we need a dedicated GPU inside of, like, a trimmer. And we’re like, okay, yeah, that sounds interesting. I was like, yeah, but that’s gonna cost you a lot. I was like, okay, how much are we talking? And it’s like, I gave him the numbers, and he’s like, okay, wow.
63 00:10:19.290 ⇒ 00:10:35.390 Dan Hartley: these… this is insane. So, is there a way you can just, like, bypass this? I’m like, if you want to just, like, have this inside of it, do it automatically? No. So, you gotta have that integrated GP right over there, and then that is how we got it. So the client said, okay, I do feel like, but I feel like,
64 00:10:35.620 ⇒ 00:10:52.060 Dan Hartley: And there was, like, a little bit of drift initially, because when I confirmed it, I was just, like, wanting to just, like, convince them for just, like, go by the idea of a mobile application first, because obviously, when I’m working on a product or something like that, I invest myself completely to make sure that I am also an evident part of the product itself.
65 00:10:52.060 ⇒ 00:10:55.790 Dan Hartley: So that is the way I can best resonate with the product itself.
66 00:10:55.790 ⇒ 00:11:04.970 Dan Hartley: So, the client went like, okay, yeah, okay, so let… I agree with you, I’m gonna drop off the idea of the trimmer, let’s go with the mobile application. And then, yeah, that’s how it proceeded.
67 00:11:04.970 ⇒ 00:11:12.989 Samuel Roberts: Okay, cool, alright, good, I like that. Yeah, I’ve had… not quite the same, but similar kind of, like, people being very,
68 00:11:13.320 ⇒ 00:11:33.020 Samuel Roberts: optimistic with what they think they can get done in a certain… yeah, of course, okay. Cool. Alright, let’s, let’s jump a little bit. Let’s talk about the, like, tooling and the models coming out, frameworks coming out, like, there’s always something new dropping. And it’s interesting, at Brainforge, we have, you know, client work, and we’re…
69 00:11:33.020 ⇒ 00:11:50.360 Samuel Roberts: maybe a little more careful what we push out to clients in terms of tools we’re using, but we’re experimenting internally with… we’re building stuff, we’re experimenting. So, I guess, tell me if there’s an example you can think of of a trend, or any of the other things I mentioned, model, framework, whatever, that you were excited about, but decided not to adopt for some reason, perhaps.
70 00:11:51.340 ⇒ 00:12:16.279 Dan Hartley: Okay, that’s a great question, that’s a great question. So, basically, obviously, you know, with how AI is changing each and every day, the entire ecosystem is pretty intense right now, right? So, a couple of years ago, it was just, like, a single prompt, and then we would be used to get entire responses, right? And then now it’s just, like, a multi-agent step reasoning, React agents where we’re going forward with, right? So, what I personally like to do is just, like.
71 00:12:16.280 ⇒ 00:12:21.480 Dan Hartley: research a lot, and then experiment internally. What I kind of do is
72 00:12:21.480 ⇒ 00:12:33.839 Dan Hartley: I should… my general approach is just, like, I use… I use quicker prototyping. I try to prototype real quick, right? So, use them with a lot of new frameworks, and then I see if I scale… if I have to scale this up.
73 00:12:34.060 ⇒ 00:12:50.790 Dan Hartley: to millions and millions of users, right? How… how can I do it, right? Is the current architecture supportive towards it? How it’s gonna be, right? So that is how I go by it on understanding how do I just, like, shift the trend, how do I just, like, use it or scale it up?
74 00:12:50.830 ⇒ 00:13:08.619 Dan Hartley: there was a time where I was working on a voice agent, and then the client wanted to just, like, have it on a very low-cost solution, and it was built entirely through WAPI, so what I did was, okay, so it’s just like drag-and-drop workflow automation, that’s how I just, like, started off, and then it was like, okay, can we scale it to make…
75 00:13:08.880 ⇒ 00:13:28.480 Dan Hartley: nearly 50,000 calls a day, and I was like, okay, that’s a tough number that we’re talking about, right? So, then I shifted it back towards a more custom architecture, a more custom, robust pipeline, and introducing some new architectures right over there. So that is how I just, like, trended it. So for me, it’s scalability, optimization, and cost.
76 00:13:30.040 ⇒ 00:13:41.449 Samuel Roberts: Cool, alright, good, good. I guess, let’s jump a little bit, because we have a little more time, I want to… I don’t wanna… I don’t want to take away your question time, but I’m curious, like, so… so we’re… I come from, like, the web dev background.
77 00:13:41.450 ⇒ 00:14:03.019 Samuel Roberts: Very TypeScript-heavy, full stack, like, just one language all the way through. We have a decent amount of Python we use. The data side especially uses a lot more Python, but we’re not afraid to use the right tool for the job kind of thing. So I’m curious, what is your tech stack of choice? Like, where do you feel comfortable? I know things change with AI, where I feel more comfortable with Python than I probably used to, or should, but I’m curious where you feel.
78 00:14:03.270 ⇒ 00:14:19.230 Dan Hartley: So, I’m pretty open towards just, like, I believe, like, at this point, I’m a kind of person who can just, like, adapt to style. I’m the most comfortable with Python. I’ve been working with Python, I’ve been working with LineChain because of the… because of the advancements in AI, and it’s…
79 00:14:19.230 ⇒ 00:14:35.929 Dan Hartley: diversity to just, like, have it all done. But, I work with TypeScript, I’ve, I’ve worked with JavaScript, I’ve worked with React, I’ve worked with Node, not necessarily… and it was way back when I was just, like, doing the software engineering kind of thing, but over the past couple of years, I’ve been more Python-focused.
80 00:14:36.430 ⇒ 00:14:40.789 Samuel Roberts: Cool, okay, good to know, good to know. Alright, so yeah, I guess I want to kind of change
81 00:14:40.900 ⇒ 00:14:50.500 Samuel Roberts: change direction, and whatever questions you have, about the role, about the company, about whatever I can kind of talk about, client-wise, but yeah, anything I can answer.
82 00:14:50.710 ⇒ 00:15:07.269 Dan Hartley: Definitely, definitely. So, I just want to just, like, start off. So, I actually have a couple of questions with Kara, in the story, and then I just, like, wanted to just, like, know it on a deeper level on, what… at Brainforge, right? So, what is the most toughest challenge, technology-wise, that you guys are facing right now?
83 00:15:08.450 ⇒ 00:15:11.149 Samuel Roberts: Technology-wise, what is the toughest?
84 00:15:11.800 ⇒ 00:15:18.359 Dan Hartley: In terms of the technical capabilities, in terms of what, are we fixing any bottlenecks, are there any bottlenecks, something like that.
85 00:15:19.260 ⇒ 00:15:41.479 Samuel Roberts: Yeah, I mean, we’re… we’re kind of… what I… I mean, I don’t necessarily think that technology is the biggest bottleneck right now for us. You know, I think, like, we’re… I mean, obviously, like, more hands on deck, like, more people is always helpful, and so, like, we’re… we’re looking to grow and hire, but I would say, like, the big thing technologically is just understanding how best to use, and, we’re really working on a lot of the coding agent stuff.
86 00:15:41.480 ⇒ 00:15:49.070 Samuel Roberts: So we’re, we’re all in Cursor. We actually got, like, most of the company, even non-technical people, into Cursor, because it’s been…
87 00:15:49.070 ⇒ 00:16:07.070 Samuel Roberts: you know, we have all our Markdown files, all the meetings and everything. We’ve built a little bit of a platform on the web, but to get all the agentic, you know, conversations over that, have it skills and all this other stuff, we’ve really been leaning into Cursor. And so I think the next big step for us, like, as a company, is figuring out, is that the right
88 00:16:07.100 ⇒ 00:16:28.500 Samuel Roberts: form factor, it’s just what’s available now. We’re really experimenting with that a lot, from, like, an operational standpoint. From the building side of things, I think our biggest thing is just figuring out, for clients, you know, they come in, they have an idea, they think they can, oh, I need this AI tool, and really it’s digging in to say, okay, what’s the problem?
89 00:16:28.500 ⇒ 00:16:33.900 Samuel Roberts: you know, they think they want this, but really, this is better. Like, that kind of stuff, I think, you know, the technology is almost becoming…
90 00:16:33.950 ⇒ 00:16:42.920 Samuel Roberts: like, so fast. It’s like, you can throw together whatever and see if it works kind of thing pretty quickly at this point. You know, scalability is a…
91 00:16:43.080 ⇒ 00:16:45.620 Samuel Roberts: Not as big a concern, because we’re building, kind of.
92 00:16:45.720 ⇒ 00:17:02.110 Samuel Roberts: tools for clients rather than, like, you know, hyperscale kind of things. But I think the biggest thing is really just how… how do we build… what… what technology we use to build more technology kind of thing is really the… the craziest thing right now for us. So that’s, yeah.
93 00:17:02.370 ⇒ 00:17:13.030 Dan Hartley: Got it, yeah, I think that’s really exciting, because obviously you gotta work, on so many different, different… so one day you’re working on a single technology, and the other day you gotta work, just, like, adapt to it, and be just, like, fast-paced, just like an.
94 00:17:13.030 ⇒ 00:17:13.540 Samuel Roberts: Yes.
95 00:17:13.540 ⇒ 00:17:18.159 Dan Hartley: So I know it gets crazy sometimes, and it just, like, excites me a lot.
96 00:17:18.160 ⇒ 00:17:19.260 Samuel Roberts: Yeah, great.
97 00:17:19.260 ⇒ 00:17:30.129 Dan Hartley: Yeah, so I want to know about the role itself. So, what does success look like for this specific role? How does a specific day-to-day look like at the Brainforge, being a machine learning engineer, and yeah.
98 00:17:30.880 ⇒ 00:17:32.470 Samuel Roberts: Yeah, so,
99 00:17:32.760 ⇒ 00:17:49.589 Samuel Roberts: let me just give you a little bit of context there. So, Brainforge started as a data consultancy, so we have kind of the data team on the engineering side, and then we have the AI team. And so that kind of spun out of a lot of the internal work that was happening, clients asking about it, so now we’re kind of building that out a little bit more. And so,
100 00:17:50.200 ⇒ 00:18:02.490 Samuel Roberts: Excuse me. So, we kind of have, like I said, two different sides of it. There’s the client work, various clients, different projects, different levels of complexity, different scopes in general. And then there’s the internal work, which is…
101 00:18:02.490 ⇒ 00:18:18.719 Samuel Roberts: kind of the stuff I talked about of figuring out how we’re doing things, building, you know, tools for the other parts of the team, marketing, sales, things like that. So, the way the AI team kind of works is we try to think of the internal stuff as a client, and so allocation-wise, like, you might be on
102 00:18:18.720 ⇒ 00:18:24.160 Samuel Roberts: You know, a client or two, depending on how deep you are into their project, and then maybe doing some internal stuff.
103 00:18:24.160 ⇒ 00:18:39.730 Samuel Roberts: It also varies by, you know, what’s currently the work we have. You know, clients churn, clients sign up, clients expand scope, reduce scope, so, like, things like that tweak a little bit, but day-to-day, I mean, what I think, you know, the biggest thing for this role is, you know, we’re all remote.
104 00:18:39.730 ⇒ 00:18:51.570 Samuel Roberts: We’re somewhat async, you know, we’re on Slack, we’re trying to limit how many face-to-face meetings and stuff that I just… each time, more than anything. But really just, like, being able to execute.
105 00:18:51.570 ⇒ 00:18:54.849 Samuel Roberts: You know, come together, discuss a plan, refine the plan.
106 00:18:54.900 ⇒ 00:19:04.710 Samuel Roberts: And then execute. Especially with a lot of the coding agents now, like, the planning… I’m trying to shift a lot of our focus towards not just, like, oh, here, yeah, this is the problem, go run with it, but, like.
107 00:19:04.710 ⇒ 00:19:23.789 Samuel Roberts: here’s how you might want to think about it, here’s what you might want to feed to Cursor, or better yet, go chat with Cursor, come back, and we’ll discuss it as a team. Or if it’s something simple, like, yeah, just go do it and update the ticket kind of thing. But really, I mean, that’s what we’re looking to do, is just execute, keep our, you know, our velocity pretty good on tickets,
108 00:19:24.080 ⇒ 00:19:26.840 Samuel Roberts: Excuse me. So yeah, I mean, you’ll be in different…
109 00:19:26.930 ⇒ 00:19:42.170 Samuel Roberts: client projects, and it’s hard to really say what that is. It’s a little different, I feel like, than the data side, where we have a little more defined service lines, and AI is a little more product-y right now. AI automation is kind of what we call it, so, like, there’s some things that are more…
110 00:19:42.170 ⇒ 00:19:53.440 Samuel Roberts: you know, chat rag standard things. We’re trying to… we have one client that we’ve been working with for a while, where we put together something, we’re kind of migrating that to some new tooling now, and it’s…
111 00:19:53.900 ⇒ 00:20:09.630 Samuel Roberts: it’s a whole thing, like, the data quality and, like, lots of sort of that stuff. Whereas, like, we’ve had some other clients that was just like, hey, I do this in Cloud, I paste this over and over and over again, and we just built them a simple automation. Other, you know, we’re always looking for kind of new and exciting things that way, but it’s less of a…
112 00:20:09.630 ⇒ 00:20:17.959 Samuel Roberts: like, one-size-fits-all sort of thing. So, lots of different stuff, we’re always trying to think of new things, we’re always playing with new things internally that we can bring externally.
113 00:20:18.510 ⇒ 00:20:20.360 Samuel Roberts: But yeah, that’s kind of the main…
114 00:20:20.360 ⇒ 00:20:24.969 Dan Hartley: Definitely, I mean, like, really, thank you so much for just, like, explaining it in that depth and detail.
115 00:20:24.970 ⇒ 00:20:25.300 Samuel Roberts: Totally.
116 00:20:25.900 ⇒ 00:20:42.330 Dan Hartley: Yeah, and I think that’s a good initiative on everybody using Kirscher, but using it to get… give it the right context first, because even when I use Kersher, or even when I use Windsurf, I try to just, like, make sure that I am building the right context first, because that’s important. Else, it’s just going to take me…
117 00:20:42.480 ⇒ 00:20:53.749 Dan Hartley: 2 hours to just, like, code a feature out, and 20 hours to just, like, debug it out. So, being an AI engineer, you know, just, like, how involved at least, like, coding is gonna get. So yeah, I understand, I understand that part, yeah.
118 00:20:53.750 ⇒ 00:20:57.810 Samuel Roberts: Yeah, yeah, it’s, it’s, it’s… it’s a really interesting problem from, like, the…
119 00:20:58.030 ⇒ 00:21:15.529 Samuel Roberts: the low level of, like, day-to-day what you’re doing, but also trying to think, like, how do we get the company, how do we get everyone having the right context? Like, we’ve made a few, like, interesting decisions that way, trying to see how it goes, and I think a lot’s gonna evolve in the next year, and probably longer, you know what I mean? Like, I feel like we’re in a very…
120 00:21:15.540 ⇒ 00:21:21.400 Samuel Roberts: influx time in the tech world in general, and so I’m curious to see where some of it shakes out, and we’re kind of on the…
121 00:21:21.590 ⇒ 00:21:27.270 Samuel Roberts: like, cutting edge of using that stuff and really trying to build, like, an AI-first consultancy and,
122 00:21:27.810 ⇒ 00:21:28.850 Samuel Roberts: You know, yeah, it’s cool.
123 00:21:28.850 ⇒ 00:21:38.160 Dan Hartley: And that answers my last question, because my last question was about to be that how adaptable are we guys here at Brain Forge? And that completely answers it, because
124 00:21:38.830 ⇒ 00:21:54.070 Dan Hartley: the way we’re adapting towards how we’re progressing with the new advancements in AI, making it more AI-focused first, and that’s how the trend is going to be, and over the years, it’s gonna just, like, increase and expand. So, I’m like, yeah, it’s great to see. So I think, yeah, that’s pretty much it that I wanted to ask at this point.
125 00:21:54.290 ⇒ 00:22:03.920 Samuel Roberts: Yeah, no, I mean, just to build on that a little bit, like, when I… I came from a startup background, more product-focused, so joining the consultancy was kind of weird for me to begin with, but then, let alone,
126 00:22:04.260 ⇒ 00:22:06.270 Samuel Roberts: how AI-forward
127 00:22:06.800 ⇒ 00:22:14.740 Samuel Roberts: thinking the whole company is, and especially from the top down with Utena’s CEO, like, he’s very, you know, try it first. Like, see if the AI can help you. See, like.
128 00:22:14.740 ⇒ 00:22:35.370 Samuel Roberts: Can you increase your velocity? Can you do this? Can you do that? Like, where are the limitations? And so, I think there’s something to be said for that. Like, a lot of companies are probably trying to figure out… and we’re working with companies that are literally just like, we need AI, it came from the top down. Someone said, use AI, and we don’t really know what to do. And so, we’re trying to not only help them, but do it for ourselves in a very forward way.
129 00:22:35.670 ⇒ 00:22:48.079 Dan Hartley: Yeah, and I think that’s nice. Yeah, I recently saw a meme over the internet over the past couple of days of this, like, big startup and big companies, they were like, okay, what do we need right now? We need AI. Okay, so why do we want it? We don’t know. When do we want it?
130 00:22:48.080 ⇒ 00:22:48.430 Samuel Roberts: Yeah.
131 00:22:48.430 ⇒ 00:22:52.519 Dan Hartley: right now, so yeah, I get it. Yeah, I get it, I get it. That’s how the notebook is.
132 00:22:52.520 ⇒ 00:22:54.790 Samuel Roberts: Definitely, definitely. Yeah, no, it’s a, it’s,
133 00:22:55.390 ⇒ 00:23:14.640 Samuel Roberts: it’s crazy, like, what we’re doing that we weren’t doing a month ago, two months ago, six months ago, what people are… what we’re trying to figure out to sell to clients, and, you know, what new things can we do that way, how much better are the models getting, how much better is the tooling getting? There’s just… there’s so much changing, and it’s a really exciting time to be working on this sort of stuff.
134 00:23:15.010 ⇒ 00:23:23.530 Dan Hartley: Definitely, definitely. Likewise, yeah. So, yeah, these were the pretty end-up questions that I wanted to ask from my end. I want to leave it more back to you if you have something else to ask.
135 00:23:23.620 ⇒ 00:23:44.620 Samuel Roberts: Yeah, I mean, we’ve covered pretty much everything, like, I try not to get too technical on these, like, I want maybe a little… maybe we can dive a little bit more, since we have a little bit of time, and it might just, you know, help the next interviewer, but, so you mentioned Langchain, Python, as kind of the main things, so, but you also mentioned, like, you’ve done a little bit of, like, maybe front-end React stuff. How comfortable do you feel, like.
136 00:23:44.620 ⇒ 00:23:46.790 Samuel Roberts: You know, I guess my question maybe is.
137 00:23:46.790 ⇒ 00:23:50.819 Samuel Roberts: for the projects where you’re building stuff in Python, what is the, like, user…
138 00:23:50.820 ⇒ 00:23:57.550 Samuel Roberts: end look like? Are you providing a web interface, or what, like, what are you doing for that? Just talk me through that a little bit, just so I can…
139 00:23:57.550 ⇒ 00:24:03.020 Dan Hartley: Yeah, so, yeah, so for… it basically depends. There’s, like, an entire front-end team that handles the front-end side of things, and then.
140 00:24:03.020 ⇒ 00:24:03.870 Samuel Roberts: Okay, okay.
141 00:24:03.870 ⇒ 00:24:21.120 Dan Hartley: side of things and done everything, but I am, being a lead, but obviously, I know, I gotta stay up-to-date on just, like, what’s happening around. So, on the front side of things, yeah, it’s majorly React to build out, and it depends if it’s, like, web pages for itself, like, React, if it’s a mobile application, we go by React Native.
142 00:24:21.120 ⇒ 00:24:29.850 Dan Hartley: And, yeah, that’s pretty much it. In terms of comfortability, I am aware of the basic concepts on how they’re working, you know, and then…
143 00:24:30.090 ⇒ 00:24:33.669 Dan Hartley: Yeah, on how the components should render, each and everything like that.
144 00:24:33.670 ⇒ 00:24:34.390 Samuel Roberts: For sure.
145 00:24:34.970 ⇒ 00:24:44.760 Samuel Roberts: So, okay, let’s… okay, that’s interesting, so I’m just thinking a little bit here. So, like, what… maybe there’s another question I should be adding sort of earlier in the process, but, like, what’s, why are you looking to…
146 00:24:44.880 ⇒ 00:24:47.900 Samuel Roberts: you know, not just come to Brainforge, but leave your current role, kind of thing.
147 00:24:47.900 ⇒ 00:25:12.279 Dan Hartley: That’s… that’s a great question. So, in my current role, obviously, I mentioned this to Kara, that almost all of the projects, they are in the maintenance space right now. And obviously, when I’m being… working as a tech lead, there’s, like, almost… there’s, like, a lot of juniors, and I completely respect that they have a lot of features to learn, their own self as well. So as soon as a feature hops in, I create tickets out, and they be, like, they go all excited that, hey, I want to take on this ticket, I want to take on this ticket.
148 00:25:12.280 ⇒ 00:25:12.820 Samuel Roberts: Right.
149 00:25:13.390 ⇒ 00:25:30.619 Dan Hartley: So I feel like, okay, I’ve seen it for the couple of weeks, and I feel like, okay, I think it’s the right time to step back, and it’s the right time. I think I’ve reached the ceiling. And when I joined them, I was one of the foundational engineers in the AI space, right? So I feel like, okay, I can go back, relive the experience again.
150 00:25:30.620 ⇒ 00:25:36.660 Dan Hartley: Build… grow together, and scale up with another firm, and then use my… utilize my skill upside down.
151 00:25:37.560 ⇒ 00:25:44.799 Samuel Roberts: Great, yeah, no, I… I feel that I’m… I’m struggling with how much to be in code versus not sometimes, you know.
152 00:25:44.800 ⇒ 00:26:02.770 Samuel Roberts: delegating and, like, I come from, like I said, a startup background where it was, like, me and, like, one or two other guys, kind of thing. Maybe a designer, and another engineer, or a business… you know, so, like, I’m used to, I just do it myself, and now the AI makes me feel like, oh, no, I can do everything, and making sure that, like, the team has the stuff to do, but also, like, I don’t want to stop learning is an interesting.
153 00:26:02.770 ⇒ 00:26:03.220 Dan Hartley: Yeah.
154 00:26:03.220 ⇒ 00:26:06.429 Samuel Roberts: kind of worry, so I understand that. Cool.
155 00:26:08.020 ⇒ 00:26:16.290 Samuel Roberts: Let me just make sure I hit everything I wanted to here… Yeah, I think we’ve covered pretty much everything I’ve got here. So, I kind of mentioned the process, but yeah, I’ll obviously bring this back to the team.
156 00:26:16.610 ⇒ 00:26:25.940 Samuel Roberts: And then the next step would be a more kind of technical role-focused, and then a tech challenge and a panel interview, and then an offer, and we…
157 00:26:26.230 ⇒ 00:26:42.410 Samuel Roberts: I don’t know the exact timeline on all that, but I keep saying we’re moving relatively quickly, we’re… we don’t want to drag things out, you know, you’ll hear one way or another, either no or yes, schedule this, and then the scheduling is probably the part that takes the longest sort of thing, just to make sure that, like, that works out. And then,
158 00:26:43.390 ⇒ 00:26:50.790 Samuel Roberts: yeah, that’s the whole process at that point. So if you have any questions, feel free to reach out. But that’s it, yeah, thank you so much for your time.
159 00:26:51.060 ⇒ 00:26:57.350 Dan Hartley: Likewise, Samuel. It was great speaking, I hope you talk again soon, and yeah, definitely looking forward towards the next steps!
160 00:26:58.090 ⇒ 00:26:58.750 Samuel Roberts: Great.
161 00:26:58.890 ⇒ 00:26:59.800 Samuel Roberts: Have a good one.
162 00:26:59.800 ⇒ 00:27:01.240 Dan Hartley: Thank you. Bye.
163 00:27:01.560 ⇒ 00:27:02.150 Samuel Roberts: Bye.