Meeting Title: Brainforge Interview w- Sam Date: 2026-03-11 Meeting participants: Samuel Roberts, Marcus
WEBVTT
1 00:02:42.660 ⇒ 00:02:43.470 Samuel Roberts: Hello?
2 00:02:43.580 ⇒ 00:02:44.869 Marcus: Can you hear me?
3 00:02:45.480 ⇒ 00:02:46.100 Samuel Roberts: And now.
4 00:02:48.220 ⇒ 00:02:49.019 Marcus: Can you hear me now?
5 00:02:49.440 ⇒ 00:02:50.130 Samuel Roberts: Second.
6 00:02:50.590 ⇒ 00:02:51.210 Samuel Roberts: How are you?
7 00:02:52.380 ⇒ 00:02:54.520 Marcus: And then, good, thanks for asking, what virtue.
8 00:02:54.790 ⇒ 00:02:57.639 Samuel Roberts: Yeah, doing alright, doing alright. Thanks for taking the time today.
9 00:02:58.550 ⇒ 00:02:59.130 Marcus: Great.
10 00:02:59.550 ⇒ 00:03:10.030 Samuel Roberts: So, I’ll just start with a brief intro myself, and then you can intro yourself. I have some questions, I’ll leave some time for you to ask questions, but, let’s just jump in. So my name is Sam Roberts.
11 00:03:10.270 ⇒ 00:03:15.599 Samuel Roberts: I’m the, the AI and Automations Tech Lead here at Brainforge.
12 00:03:15.810 ⇒ 00:03:20.730 Samuel Roberts: I know you spoke with Kayla previously, I believe, right?
13 00:03:21.030 ⇒ 00:03:24.739 Samuel Roberts: Right. Okay, cool. So yeah, so this is just kind of the first…
14 00:03:25.070 ⇒ 00:03:29.659 Samuel Roberts: you know, main interview, I guess, after that chat.
15 00:03:30.270 ⇒ 00:03:34.189 Samuel Roberts: So yeah, if you could just give me a brief intro to yourself, and we’ll get started.
16 00:03:35.330 ⇒ 00:03:47.750 Marcus: Sure thing. So, Marcus here, working in this industry for, like, the past 7 plus years, having experience majorly on the backend side of things, and, on, now AI, too. So,
17 00:03:47.980 ⇒ 00:03:54.880 Marcus: Yeah, I think, I started working as a backend engineer, and then, shifted towards some,
18 00:03:54.880 ⇒ 00:04:15.980 Marcus: full-stack roles, worked on two of the projects where I was working on the front end as well at the back end, and for the past two years, I’ve been heavily working with the AI agents and working automation, and basically built around 7 different applications with AI agents. So either they are basically chatbots, simple text-to-text RAG implementation, either it is text-to-SQL, either they are audio bots.
19 00:04:15.980 ⇒ 00:04:35.539 Marcus: For taking interviews. So, basically, those kind of, bots that I’ve been working with. Other than that, I’ve been, very good, working across different fields. So, I have experience working in fintech, I have experience working in the healthcare, I have experience working in real estate, and then working in, e-commerce, too.
20 00:04:35.540 ⇒ 00:04:37.280 Marcus: So, yeah, that’s a bit about me.
21 00:04:38.010 ⇒ 00:04:39.410 Samuel Roberts: Great, thank you.
22 00:04:40.260 ⇒ 00:04:43.810 Samuel Roberts: Okay, so let’s just jump in. I got a few,
23 00:04:44.090 ⇒ 00:04:53.429 Samuel Roberts: questions here, and then, we’ll see where we go. So, you mentioned a few examples, but can you give me a kind of,
24 00:04:53.940 ⇒ 00:05:01.269 Samuel Roberts: Tell me about an LLM-based feature that you’ve built, shipped to production, and, like, specifically the problem that it solved.
25 00:05:01.880 ⇒ 00:05:25.900 Marcus: Sure thing. So, the most recent one was working, on an AI interviewer. So, the talk with Kayla… So, usually, any talk with recruiters is a same set of questions that they ask over 30 minutes called to every candidate, right, for that particular job. So, let’s just say… so, the product statement in our company was… so, basically, our company was working directly with another company that had a lot of job postings.
26 00:05:25.900 ⇒ 00:05:27.799 Marcus: Due to… I don’t know why.
27 00:05:27.800 ⇒ 00:05:51.539 Marcus: So they were posting more than 20 different jobs a day, so they wanted a huge man force just to cater all those interviews. That was, of course, not possible. So, what they wanted is that, why not automate this process? Why not just screen those candidates and then start moving towards the technical rounds? And basically, screening of the first rounds is very important. It takes a lot of time and a lot of manpower, but still,
28 00:05:51.940 ⇒ 00:06:16.730 Marcus: a few candidates make it towards the next steps, so that, that’s what our AI interviewer was doing. So we, the recruiter fed all the questions that they wanted to ask in the first interview, and those questions were fed to that AI bot, and that AI bot was basically whenever a candidate applied to a particular job, the AI interviewer’s link was sent to their email, and whenever they take their interview, we would be getting their transcript, we would be getting their audio and video.
29 00:06:16.730 ⇒ 00:06:24.500 Marcus: So, audio-video was just to make sure that the candidate is the candidate he says, and again, there’s not another AI talking to the AI function.
30 00:06:24.500 ⇒ 00:06:25.460 Samuel Roberts: Sure, sure.
31 00:06:25.990 ⇒ 00:06:27.180 Marcus: So these kind of…
32 00:06:27.180 ⇒ 00:06:50.059 Marcus: edge cases were being catered using audio and video, and then there was a transcript. So, firstly, we had this way of vetting the candidate. So, the recruiter, what they would do is that they would go through the transcript, they would go through a brief skimming of the audio and video, and they would make sure that it’s good, yeah, the candidate’s so good, and transcript, they were judging it via the transcript.
33 00:06:50.060 ⇒ 00:07:06.279 Marcus: of that interview. But afterwards, we also added another agent who was basically creating that transcript on the basis of the JD that we were provided. So, yeah, so there was an LLM grader to grade, that particular person’s or candidate’s answers, just to make sure how well they aligned with our particular job description.
34 00:07:06.280 ⇒ 00:07:10.490 Marcus: Yeah, this is a good product that I worked on from the front end, back end.
35 00:07:10.490 ⇒ 00:07:16.949 Samuel Roberts: Yeah, that was gonna be my next question, was tell me more about the stack, like, where have you spent the most time, and tell me if you want to go into the details about that one.
36 00:07:17.360 ⇒ 00:07:30.429 Marcus: Yeah, sure. So, this was the first time me working in TypeScript, so I had never… I didn’t have any experience working with TypeScript. Next.js, so they were leveraging Next.js throughout, so client-side and server-side, so…
37 00:07:30.430 ⇒ 00:07:54.650 Marcus: Basically, that was the entire, architecture of their product. So that’s why they wanted it to be native, for their… because they had MonoRepo, and they had a huge core base, due to Monorepo. They had 7 different services working in one repo. So, yeah, this thing was added in another project feature, product in their, yeah, in their codebase.
38 00:07:54.650 ⇒ 00:08:04.960 Marcus: So, yeah, that’s what I leveraged. I’m not… I won’t say I’m still very good with TypeScript, but I think, Python is what I always prefer, but yeah, I had to work on TypeScript, so…
39 00:08:05.830 ⇒ 00:08:17.530 Samuel Roberts: Okay, so that was a full stack, or next project, what was the, like, what were you using to interface with the LLM? Like, what was the… were you just making direct calls, or were you using a framework? OpenAI? Yeah, OpenAI.
40 00:08:17.530 ⇒ 00:08:21.909 Marcus: Okay. So they have, what is called,
41 00:08:23.120 ⇒ 00:08:27.089 Marcus: there’s an API, a real-time API, it’s called real-time API.
42 00:08:27.090 ⇒ 00:08:27.440 Samuel Roberts: Oh, sure.
43 00:08:27.440 ⇒ 00:08:27.880 Marcus: For the…
44 00:08:27.880 ⇒ 00:08:35.500 Samuel Roberts: Okay, yep, we’ve used that for a few things. Yeah, okay. Cool, cool. All right, great. Okay, so let’s talk a little bit about…
45 00:08:37.270 ⇒ 00:08:56.689 Samuel Roberts: when you’re dealing with stakeholders, you know, sometimes they’re non-technical, and there’s a lot of stuff out there about AI and LLMs, and they have, you know… how do you… how would you go about, or have you gone about, explaining, like, the limitations, maybe, that they might not realize are there, especially if they’re non-technical?
46 00:08:57.600 ⇒ 00:09:11.940 Marcus: So, it’s around 3 times that I have faced this issue, so I don’t know, it’s coincidence or what. So, every time I’m getting assigned to a new project, it is to a person who is directly a non-tech person, and either
47 00:09:11.940 ⇒ 00:09:31.779 Marcus: or either he’s just from the upper management. So, I directly work with those. So, what I do is that I basically make sure in our conversations, whenever they are laying out their user stories or use cases of their product, what they have in mind, I make sure that they are not over-exaggerating or over,
48 00:09:31.780 ⇒ 00:09:35.969 Marcus: empowering the AI that we currently have, so sometimes they want
49 00:09:35.970 ⇒ 00:09:59.319 Marcus: cognition out of the AI, which means that they want AI to learn things out of their own, they want it to outperform… they should learn… they want the AI to learn from itself, from its mistakes, and then just answer it perfectly on the other go. Basically, that’s what kind of questions that I try to highlight whenever we talk about any AI
50 00:09:59.320 ⇒ 00:10:07.059 Marcus: not just LLMs or agents. Whenever we talk about any model, we make sure… I make sure that they know that the data is really important.
51 00:10:07.060 ⇒ 00:10:26.190 Marcus: the AI would be only answerable to the questions that that data is provided off, not any other thing. Basically, that would be hallucination of our AI. Just like if you provide just the services company data and, the company, or let’s just say our AI interviewer, I’m providing 10 questions to that AI to get answers of.
52 00:10:26.190 ⇒ 00:10:28.880 Marcus: It won’t be able to answer, basically.
53 00:10:28.880 ⇒ 00:10:29.240 Samuel Roberts: Right.
54 00:10:29.240 ⇒ 00:10:34.620 Marcus: ask those questions that are not provided. So basically, those kind of limitations is what I try to,
55 00:10:35.180 ⇒ 00:11:00.069 Marcus: explain whenever we are working on the requirements gathering, whenever we are working on the use cases. So, as soon as possible, I try to, communicate those. If that’s not possible in our call, just whenever they are talking about, when I always take some time to do some research whenever… so I, usually say, give me 2 or 3 hours just to do some research and make sure that we are on the same page, and I do not commit anything that’s
56 00:11:00.070 ⇒ 00:11:02.630 Marcus: It’s not possible, or that’s out of our bounds.
57 00:11:02.630 ⇒ 00:11:25.969 Marcus: So, yeah, these kind of things. Another… the last one would be not throwing technical jargons on them. So, if you are a technical person, I have told you about the real-time API in Next.js. I won’t be telling them all of that. I’m just saying that AI, open AI, is what I am leveraging. ChatGPT is what I’m leveraging, and yeah, these kind of stuff is what they understand, this is what they relate to, and other than that,
58 00:11:25.970 ⇒ 00:11:35.710 Marcus: working on demos. I think the unanimous tone of a developer and a non-tech person is the UI, the front end, the…
59 00:11:35.740 ⇒ 00:11:56.469 Marcus: visuals of any product. So, sometimes there are gaps where they can’t really explain the requirements, and where we can’t really get their requirements, but when there’s an interface, they know that this is not what we require. We require a button not here, we require a button here, or the functionality should be like this. Basically, that’s where I try to connect with them.
60 00:11:56.920 ⇒ 00:12:06.279 Samuel Roberts: Okay, great. So is there, is there a time you can tell me about where someone did misunderstand the feature and what it could do, and how you resolved that with them?
61 00:12:07.440 ⇒ 00:12:08.190 Marcus: So…
62 00:12:08.450 ⇒ 00:12:18.700 Marcus: Yeah, I think, there was one time where, where the person, basically the stakeholder, I was working on a fintech platform with the product.
63 00:12:18.700 ⇒ 00:12:41.769 Marcus: that they wanted to talk to their data. So, they had huge investments done, the person himself, the CEO himself, and his colleagues, they had a huge, database of investments that they did for the past two decades. So, they wanted to talk to the data. So, the issue was that he wanted, so, the, the product that I built is what he wanted. So they were, he wanted, basically.
64 00:12:41.770 ⇒ 00:12:50.679 Marcus: that if I type in any question, that who were the 10… top 10 performing efforts of mine for the past 10 years? So basically, those kind of questions were, basically.
65 00:12:51.060 ⇒ 00:12:58.939 Marcus: are possible to ask out of the LLM based out of the data that we had. But he was expecting,
66 00:12:58.940 ⇒ 00:13:16.790 Marcus: a lot out of that AI. I think he was expecting the graphs to be generated out of their own. He wanted to know that why this asset performed well, and why not. So, basically, there were two factors which were in our database, which were basically directly affecting
67 00:13:16.790 ⇒ 00:13:41.270 Marcus: the boom or the, the downfall of a particular asset, but they were not sufficient to basically make, make a decision out of it. So I made sure that he knew, but, he didn’t, he didn’t understand. He was basically, he was, he wanted… he saw that the data is complete, he basically assumed that the data is complete. I made sure that, he knew after… so.
68 00:13:41.290 ⇒ 00:13:54.940 Marcus: I wasn’t, able to communicate this perfectly on the time of discussion, that the data isn’t complete, but when I completed it, I told him that, just, to make sure that we are on the same page, I’ll,
69 00:13:54.940 ⇒ 00:14:11.009 Marcus: complete this product, complete this feature, and you’ll know what the limitations are. And once he toggled, or once he played with that particular feature, he knew that, yeah, exactly, these are just two features that the assets are basically relying on, and these are not sufficient. So basically, that’s the scenario that I worked.
70 00:14:11.630 ⇒ 00:14:13.550 Samuel Roberts: Great, thank you.
71 00:14:14.030 ⇒ 00:14:25.780 Samuel Roberts: Okay, cool. Let’s, let’s change a little bit here. So let’s talk about the, kind of, industry in general. Lots of things are coming out all the time, you know, different models, different frameworks, different tools,
72 00:14:26.150 ⇒ 00:14:36.370 Samuel Roberts: what’s, like, a trend or something that’s come out that you were initially excited about but decided not to adopt for some reason? You know, I’m curious about your thoughts there.
73 00:14:37.420 ⇒ 00:14:56.559 Marcus: So, it would be working. So, I was excited whenever a new agent or a new coding tool comes out, so I get really excited, I try to play with it, I try to do a lot of stuff with it, but I tend to use them as least as possible. The reason is.
74 00:14:56.870 ⇒ 00:15:21.539 Marcus: Yeah, so the reason is, so, whenever you rely too much on these agents, so they basically lead you to a position where the product goes to a deadlock situation, the product doesn’t remain so much, extendable, scalable, so this is, what I have faced, myself in a particular position, that our… we were leveraging too much, everyone was pushing forward
75 00:15:21.540 ⇒ 00:15:38.990 Marcus: PRs, 5 PRs a day. They were raising four, five PRs a day. Everyone, there were 7 members in the team. There were 30 PRs a day, almost 20 to 30 PRs raised a day. So that product, so there was, and, basically, the customer side of, we had a huge number of bugs.
76 00:15:38.990 ⇒ 00:15:54.020 Marcus: raised out of it. The reason is you don’t… whenever… whenever you don’t code yourself, you basically oversee things, or you undersee things, I would say. So there are a lot of out… yeah, inline things that you do not see, and that’s where the product doesn’t…
77 00:15:54.760 ⇒ 00:16:17.699 Marcus: grow to a stage that you want it to be. So, I always try to use, these agents just for the planning purposes. I try to brainstorm my solution first. I do not just go and, yeah, I want to build an AI interviewer, can you please write the code? I won’t be doing exactly that. So, I just go and ask, give it my complete problem first. I try to brainstorm, I try to weigh the pros and cons of
78 00:16:17.700 ⇒ 00:16:29.040 Marcus: the solution they provided. Of course, I do have my mind of my own… a mind of my own. I provide solutions of mine, and basically, I try to make sure that
79 00:16:29.040 ⇒ 00:16:42.029 Marcus: why my solution is not good, and why the solution proposed by that AI is good. So basically, these kind of questions is what I try to do, and I’m not completely relying on, its cognition or its,
80 00:16:42.030 ⇒ 00:16:58.329 Marcus: code, I just try to make sure that we have our gears on, we have the guardrails whenever I deal with this AI code. So, yeah, that’s what I would say. I am really reluctant. So, I use them a lot, but I use them with real gear.
81 00:16:58.760 ⇒ 00:17:06.040 Samuel Roberts: Okay, good, good. Okay, last thing, and then we’ll jump to your questions, but, when you’re evaluating
82 00:17:06.140 ⇒ 00:17:24.079 Samuel Roberts: a model or a framework or something to use in a project, what… what do you look for, and how do you determine if it’s production ready? So, for example, like, we have a lot of internal stuff that we do, where we experiment a lot more with tools, but then there’s the client work where we need to be a little more certain it’s production ready. So how do you draw that line yourself?
83 00:17:24.990 ⇒ 00:17:28.160 Marcus: So, again, it would be brainstorming
84 00:17:28.160 ⇒ 00:17:52.999 Marcus: Either with the agent itself, either with the team. So, whenever I have anything in mind, I try to communicate that. If there’s, if there’s a possibility that we can connect as a team, usually there are stand-ups where you, whenever you’re stuck, whenever you get stuck on anything, you discuss the solutions out loud, you get to, basically discuss that, you, basically brainstorm those
85 00:17:53.000 ⇒ 00:17:58.999 Marcus: solutions with the team, so that you are… aren’t just the only person working it. There are five or four
86 00:17:59.000 ⇒ 00:18:23.789 Marcus: any number of different minds working on that. So, that’s where, I think, for me, it works really well. Whenever I have to make a decision out of it, I, try to talk. If that’s not possible, again, I would be going back to AI agents, if that’s possible, based on the solution out of those. If not, I would be going to Google, searching out if this product or this feature is
87 00:18:23.790 ⇒ 00:18:47.500 Marcus: already built, how was it built? Or which particular language, or which particular stack, or which particular way, or which particular format, design pattern, anything, was leveraged for this particular thing. So that’s how I make sure. So just like, there was one case when there were remote containerization service that we needed for our different Docker images, and they suggested Fly.
88 00:18:47.500 ⇒ 00:18:57.429 Marcus: The reason is Fly was giving the enterprise a huge number of credits for free, so that’s why they wanted to move towards Fly. But I mentioned that Fly has this limitation of
89 00:18:57.440 ⇒ 00:19:08.330 Marcus: image size of a Docker… Docker’s image size, so it… it cannot exceed to… from 8 GBs. So, we… I knew that we had huge, Docker images out of
90 00:19:08.330 ⇒ 00:19:29.570 Marcus: 12, 15, 17 GBs of image size, so, but, of course, they had, free credits, so they wanted it to pursue, but, there were huge downsides, on whenever we had this, Docker image on the fly. The reason is it would exceed the limit, and then we have to completely shift from fly to ECS.
91 00:19:29.570 ⇒ 00:19:35.179 Marcus: elastic containerization service of AWS, and that was a huge shift. The reason is, whenever you
92 00:19:35.330 ⇒ 00:19:39.559 Marcus: do not make the right decision on the infrastructure level. It basically…
93 00:19:39.870 ⇒ 00:20:01.410 Marcus: disrupts everything, either on the client side, either on the technical side. Everyone on the technical side gets feared up, I would say, or, yeah, and they have, to meet fit deadlines, just to make sure that the client doesn’t face the downsides. So yeah, these kind of questions is what I try to, leverage, and try to weigh the pros and cons, and just talk.
94 00:20:02.150 ⇒ 00:20:11.600 Samuel Roberts: Perfect, thank you. Yeah, so we’re a little over halfway. I want to leave some time for you to ask whatever questions you have of me, of the role, the company, whatever you’ve got, so yeah.
95 00:20:12.420 ⇒ 00:20:23.799 Marcus: Sure thing. So, starting with, I would love to know a bit more about the role itself. So yeah, it’s AI and automation role, but I would love to know a bit more about the product itself, if there’s just one product or multiple products.
96 00:20:23.800 ⇒ 00:20:29.930 Samuel Roberts: Yeah, yeah. So, Brainforge, started as a data consultancy, so we,
97 00:20:30.490 ⇒ 00:20:37.289 Samuel Roberts: started… we have kind of two teams right now, engineering-wise. So there’s the data team, they do a lot of,
98 00:20:37.700 ⇒ 00:20:50.770 Samuel Roberts: pipelines, ETLs, data modeling, all that sort of stuff. So the AI automation team, kind of spun out of the fact that we were building internal tooling for ourselves, trying to help run kind of this new
99 00:20:50.770 ⇒ 00:21:00.089 Samuel Roberts: you know, relatively young company, in the most kind of efficient way. Lots of clients wanted that kind of work, too, so then the AI team kind of started getting a little bit bigger, so that’s,
100 00:21:00.220 ⇒ 00:21:04.600 Samuel Roberts: I came in over the summer to kind of helped run the team,
101 00:21:04.660 ⇒ 00:21:24.259 Samuel Roberts: we do work internally, I think I mentioned, so we have a platform where we ingest all our meetings, and all the notes, and, there’s… we have a bunch of different things I can tell you more about what we build internally. And then externally, there’s different clients. So, at any given time, there’s, you know, a few different clients, and we…
102 00:21:25.040 ⇒ 00:21:29.659 Samuel Roberts: Looking to expand that. Oh, sorry, my cats jumped on my lap.
103 00:21:30.120 ⇒ 00:21:34.789 Samuel Roberts: So, we do different projects. It varies quite a bit. So we have some…
104 00:21:35.030 ⇒ 00:21:49.679 Samuel Roberts: examples, I guess, without getting too specific. You know, some… there’s, like, a RAG chatbot for customer service, support. So not… not customers using it, but the actual customer service representatives accessing data that they need to help the customers.
105 00:21:49.680 ⇒ 00:21:58.220 Samuel Roberts: There’s some other ones where it was more automation, but still in AI. People were copying and pasting things into Claude, and had… they had kind of a workflow we were able to automate for them.
106 00:21:58.220 ⇒ 00:22:09.959 Samuel Roberts: Some other ones are just MCP, chatting over data, pulling things from different ads. So we do a lot of different stuff. We’re, we’re, you know, kind of a little… we’re a little bit different, maybe, than the,
107 00:22:10.420 ⇒ 00:22:25.380 Samuel Roberts: Then the data team, you know, they have a lot of, like, organized, you know, things that they do. The AI team’s a little more, like, we can do a few different things, we can build different products, we’re kind of experimenting with different things internally, and kind of trying to flip that to,
108 00:22:25.830 ⇒ 00:22:32.489 Samuel Roberts: client work. So it’s, it’s a, you know, I come from the startup world, I come from a product background, I was, you know.
109 00:22:32.490 ⇒ 00:22:50.909 Samuel Roberts: building, you know, one- or two-person teams, getting something zero-to-one, kind of out the door. And so this is a little bit different for me, where it’s lots of different projects, you’re kind of context switching a little bit. So it’s not… I can’t really give you, like, this is the product we build, but, that’s kind of it. It’s an interesting environment to be in, yeah, so…
110 00:22:51.060 ⇒ 00:22:58.589 Marcus: I have been a part of a consulting firm, so I know that one day they have a ticket for a Python project, the other day they have a ticket for a C-sharp project.
111 00:22:58.590 ⇒ 00:23:19.050 Samuel Roberts: Yeah, yeah, we try to stick, so we, depending on what we’re working on, Python, obviously, is big, especially on the data side. We’ve been doing a lot more TypeScript on the AI side, because we’re building UIs, and we’re building, you know, full-stack apps. We use some N8N for certain prototyping as well, but we’re trying to move more towards the code, because, one, because…
112 00:23:19.060 ⇒ 00:23:38.569 Samuel Roberts: And it has limitations as you kind of build bigger and bigger workflows, which we started to do. And then also with the coding agents, you know, as we start… as we start to structure the code and give them more instructions, like, they can do a bit more than they can do with N8N. So it helps us, you know, build a little bit that way. So, definitely more of a TypeScript
113 00:23:38.590 ⇒ 00:23:45.780 Samuel Roberts: shop overall, but, like, Python is still valuable for us, so that’s good. Yeah, other… what else can I tell you about?
114 00:23:46.210 ⇒ 00:23:47.620 Marcus: Okay, so…
115 00:23:47.730 ⇒ 00:24:01.179 Marcus: So, the question would be, how can, let’s just say I join, the company, how can I make an impact in, let’s just say, first quarter or first half of the year? How can I make an impact within the team, within the… for the…
116 00:24:01.180 ⇒ 00:24:01.900 Samuel Roberts: customers.
117 00:24:02.620 ⇒ 00:24:12.810 Samuel Roberts: Yeah, I mean, honestly, you’d be making an impact within the first few weeks, I’m sure. It wouldn’t even be that long. So, you know, we’re a relatively small team right now. I believe there’s 4 of us that are…
118 00:24:13.190 ⇒ 00:24:22.619 Samuel Roberts: you know, AI engineers, so I’m kind of the tech lead, and there’s three, but, you know, we’re looking to add another full-time engineer, and so, you know, we’re…
119 00:24:23.180 ⇒ 00:24:40.820 Samuel Roberts: we’re… it’s an interesting environment because you will be on several different projects, right? So you’ll be getting up to speed on a bunch of those quickly. There’s also some internal stuff that we’re working on, so depending on how much, allocation there is for different clients, there’s also internal tooling, which is… it’s an interesting thing because the…
120 00:24:40.820 ⇒ 00:24:53.999 Samuel Roberts: you know, the client work is what pays the bills, so, like, that’s very important, and we have to make sure to, you know, satisfy the clients and do all that. But the internal stuff is what accelerates us doing the work. So it’s this back and forth, where no matter what you’re doing, it’s gonna be helping us out
121 00:24:54.000 ⇒ 00:25:04.259 Samuel Roberts: Some way. I think, you know, early on, you’d get up to speed on a few clients, and then probably start contributing, pulling tickets relatively quickly. There’s a little bit of, like, a…
122 00:25:04.260 ⇒ 00:25:24.640 Samuel Roberts: testing kind of probationary time period at the beginning, to make sure that everyone works together and all that jazz. But, I mean, realistically, you’d be getting into things pretty quickly. We ship relatively fast, we have kind of weekly, weekly sprints, roughly, because we want to be in communication with the client. So, you know, it’s a little bit different than when we’re building our own stuff.
123 00:25:24.640 ⇒ 00:25:42.870 Samuel Roberts: But for the client, we like to… we have meetings every week with them to check in, show them what’s going on. You know, we don’t want to go too far down several weeks sprints, and then come back to the client, and they’re, oh, well, this wasn’t what I thought, so we try to keep that iteration cycle short. So, I think you’d be contributing pretty quickly, were you to join, so…
124 00:25:44.180 ⇒ 00:25:48.710 Marcus: That… that sounds good. The last question would be,
125 00:25:49.000 ⇒ 00:25:51.600 Marcus: Talking about the autonomy that
126 00:25:51.600 ⇒ 00:26:13.030 Marcus: a person coming in would be getting on the product that they work on. So I think every person, every developer just talking about me, should be accountable for what work he does. So, he should be appraised if that work product works good. If not, basically, the feedback should be there for him, too. So, how does this work, how the communication works, how everything works?
127 00:26:13.610 ⇒ 00:26:20.980 Samuel Roberts: Yeah, yeah, so we, like I said, we’re a pretty small team, and, it’s… it’s… we have a few different projects on the AI side, so I’m kind of
128 00:26:20.980 ⇒ 00:26:45.950 Samuel Roberts: floating between them a little bit. And so, you know, the… the other engineers are kind of doing a lot more of the daily work, and we kind of talk architecture and review stuff overall. So, I think, you know, when a… if you pull a ticket for, say, adding some kind of feature, you know, we would talk through the plan for that initially, depending on the complexity, obviously. Like, if it’s something simple, and the ticket is self-explanatory, you know, we do that
129 00:26:45.950 ⇒ 00:26:59.759 Samuel Roberts: that way. And then when that work is done, you know, we review it. Ideally, there’s some tests you can show, or some, evaluations, especially for a lot of the AI modeling, or, for the models and things like that. So I think, you know, the…
130 00:26:59.900 ⇒ 00:27:16.119 Samuel Roberts: we… I mean, I think there’s always room for improvement on feedback. You know, there’s never… feedback is great. It’s hard to do sometimes, especially when you’re moving quickly. We try to make sure that there’s a cadence of, you know, monthly check-ins, you know, one-on-ones, things like that.
131 00:27:16.330 ⇒ 00:27:20.969 Samuel Roberts: We are… we’re playing around with our model of meetings, you know, so we’re fully remote.
132 00:27:20.970 ⇒ 00:27:40.439 Samuel Roberts: And so that can be a very good thing in a lot of ways, because you can be off on your own doing your work and come back, without getting distracted. But it also means that, like, we have to make the effort to check in, make sure everything’s going on, and so we’re working on that, always, different forms of that. Kind of trying to figure out the best model for that, but, you know, I think, you know, you’d be…
133 00:27:40.780 ⇒ 00:27:53.289 Samuel Roberts: talking to me on a daily basis, you know, Slack updates, we work kind of synchronously and asynchronously in that way. There’s people all over the world, so different time zones can, you know, throw things off a little bit. But, yeah, I think, you know.
134 00:27:53.900 ⇒ 00:28:09.429 Samuel Roberts: it’s one of those things that if you don’t like feedback, it might not be a great place, because you’re gonna get it no matter what. When things go well, that’s great. When things go poorly, we gotta figure out why. And so, I don’t like playing the blame game, necessarily, but I like to get to the root cause and figure that out. So, that’s kind of how we think about it.
135 00:28:10.080 ⇒ 00:28:13.550 Marcus: Sounds great, sounds great. Thank you for asking. That’s all the questions that I had.
136 00:28:13.550 ⇒ 00:28:23.439 Samuel Roberts: Okay, great! So then from here, I’ll bring this back to the team, and if we move on to the next round, there’ll be another kind of more role-focused, little more technical interview.
137 00:28:23.440 ⇒ 00:28:41.090 Samuel Roberts: And then there’s a panel after a tech challenge, and then we’ll discuss that. And then that’s kind of the whole process. After that would be an offer, if everything goes well. We like to move relatively quickly. We don’t want to drag things out. I’ve been saying the biggest hurdle is just scheduling these
138 00:28:41.090 ⇒ 00:28:48.630 Samuel Roberts: You know, synchronous calls, but besides that, we like to move relatively quickly within a week or two, so you should be hearing back one way or another pretty, pretty quickly.
139 00:28:49.430 ⇒ 00:28:50.270 Marcus: Sounds good.
140 00:28:50.510 ⇒ 00:28:52.149 Samuel Roberts: Alright, thank you so much for the time.
141 00:28:52.630 ⇒ 00:28:54.020 Marcus: Thank you. Nice meeting. Bye.
142 00:28:54.020 ⇒ 00:28:55.560 Samuel Roberts: Yeah, you as well. Bye-bye.