Meeting Title: Brainforge AI Engineer Interview Date: 2025-08-13 Meeting participants: Mustafa Raja, Pranav Narahari
WEBVTT
1 00:03:28.140 ⇒ 00:03:29.330 Mustafa Raja: Hey, how are you?
2 00:03:29.860 ⇒ 00:03:31.510 Pranav Narahari: Hey, I’m good, how are you doing?
3 00:03:31.510 ⇒ 00:03:35.909 Mustafa Raja: Yeah, doing good. Sorry, sorry for the inconvenience, I should have updated it.
4 00:03:36.710 ⇒ 00:03:38.279 Pranav Narahari: No problem, no problem.
5 00:03:39.400 ⇒ 00:03:51.690 Mustafa Raja: Yeah, so, let’s get started. So, my name is Mustafa. I am, an AI engineer, in Brainforge. I’ve been working for, almost 3 months now.
6 00:03:51.700 ⇒ 00:04:10.329 Mustafa Raja: And what we sort of do is we have our own, internal platform that we are building. So it has multiple AI agents and automations that we do for our team to help our team, be more efficient or, help, help them in their pain points.
7 00:04:10.330 ⇒ 00:04:17.029 Mustafa Raja: One sort of example is that we have, we have, an automation, that’s an AI.
8 00:04:17.110 ⇒ 00:04:28.719 Mustafa Raja: AI agent, what it does is, it looks into the meeting that we have, it would be an internal meeting or a meeting with a client, and what it does, it, it automatically, based on the
9 00:04:29.510 ⇒ 00:04:37.600 Mustafa Raja: meeting transcript, and, additional information, whatever we know about our, client.
10 00:04:38.560 ⇒ 00:04:54.849 Mustafa Raja: it gets into the context of that AI agent, and we automatically build the linear ticket for that, assign those tickets to the relevant person in the team, and it’s also able to allocate the team that it needs to be in.
11 00:04:54.960 ⇒ 00:05:18.180 Mustafa Raja: So this is an example of, what we do internally to help our, PMs, similar… we have similar automations to help our sales and our, operations, and… and so and so. And, this, we started doing this, for, you know, for our internal team, and realized that this has scope.
12 00:05:18.310 ⇒ 00:05:35.649 Mustafa Raja: we can pitch these sort of things to our clients, and so we then sort of start pitching to our clients, and then we now have a pretty good client base, that we provide AI services for, and we are growing.
13 00:05:35.770 ⇒ 00:05:41.349 Mustafa Raja: So yeah, let me know, let me know about you, what’s your background and all.
14 00:05:41.930 ⇒ 00:05:58.669 Pranav Narahari: Yeah, totally, that sounds very interesting. And Utham, too, he… I met with him this past Friday, in Austin, actually, and he… he was able to mention a few of those things as well, but I think you went into further detail, which is great. Yeah, a little bit about my background is…
15 00:05:58.670 ⇒ 00:06:02.810 Pranav Narahari: You know, I worked in, like, traditional computer science, like.
16 00:06:03.230 ⇒ 00:06:22.099 Pranav Narahari: like, roles since 2021. That’s when I graduated from, like, my bachelor’s. So 2017 and 2021, did a computer science degree. 2021 to 2020… end of 2023. Worked in, like, a large-scale,
17 00:06:22.230 ⇒ 00:06:39.389 Pranav Narahari: bank in the United States, like, in the New England area. There, I was doing a lot of, React development, but also was working closely with the cloud engineering team to automate the efforts of provisioning infrastructure in the… in AWS.
18 00:06:39.420 ⇒ 00:06:44.189 Pranav Narahari: And so, that experience was good. It was my first, like, …
19 00:06:44.510 ⇒ 00:06:47.659 Pranav Narahari: Experience of, like, working in the…
20 00:06:47.780 ⇒ 00:06:56.999 Pranav Narahari: in the industry. However, I noticed, too, that these big companies work very slowly, and they have a lot of red tape, and …
21 00:06:57.160 ⇒ 00:07:09.990 Pranav Narahari: it’s hard to grow, was my experience. And so, around that time, too, end of 2022 is when ChatGPT came out, and had a few friends that were actually
22 00:07:09.990 ⇒ 00:07:21.569 Pranav Narahari: Doing their PhDs in AI, and, they were really excited about what OpenAI had dropped, which was ChatGPT. At that point, I believe it was 3.5.
23 00:07:21.630 ⇒ 00:07:22.590 Pranav Narahari: …
24 00:07:22.970 ⇒ 00:07:36.089 Pranav Narahari: And, that was really exciting for me. I’d never used a technology like that, and I think it would have been… I was using it either, like, on the first day or within the first week. At that point, it didn’t have a ton of, …
25 00:07:36.230 ⇒ 00:07:37.270 Pranav Narahari: it wasn’t…
26 00:07:37.590 ⇒ 00:07:51.769 Pranav Narahari: immediately apparent that it could be used as, like, a huge asset for software developers, but soon thereafter, Microsoft Copilot came out, I started using that for my own personal projects, and also at hackathons.
27 00:07:51.770 ⇒ 00:07:57.810 Pranav Narahari: Within… and I was based in the Austin, Texas, like, ecosystem.
28 00:07:58.000 ⇒ 00:08:04.100 Pranav Narahari: Which is kind of how I came to connect with, with them, too.
29 00:08:04.600 ⇒ 00:08:16.939 Pranav Narahari: From there, I started working with another AI engineering agency within Austin as well, and that’s where I really was able to start working with clients directly, both in…
30 00:08:17.550 ⇒ 00:08:21.050 Pranav Narahari: In an individual… in an individual, …
31 00:08:21.610 ⇒ 00:08:33.790 Pranav Narahari: a partnership, but then also in, like, a team partnership, where I worked with, like, 2 other… or actually, 3 other AI engineers. And so, yeah, I’m super passionate about this space. I really like that.
32 00:08:33.880 ⇒ 00:08:51.150 Pranav Narahari: this is a growing space that two years ago was really small, and now it seems to be growing month over month. It’s really interesting to me to be on some of the cutting edge of this stuff, and it really makes developing
33 00:08:51.220 ⇒ 00:08:53.999 Pranav Narahari: like, more fun than ever, for me, personally.
34 00:08:54.630 ⇒ 00:09:02.089 Mustafa Raja: Yeah, so I would want to know what sort of tools are you familiar with that are related to AI engineering?
35 00:09:02.660 ⇒ 00:09:08.690 Pranav Narahari: Yeah, totally. So, some no-code tools that I’ve used is, briefly a little bit of N8N.
36 00:09:09.430 ⇒ 00:09:24.770 Pranav Narahari: I built, like, a really simple workflow which, helps me with my timesheets. So, just a personal project, I just wanted to kind of mess around with N8N. Essentially, what it would do is, take in all my GitHub commits, and then…
37 00:09:24.970 ⇒ 00:09:40.169 Pranav Narahari: parse it through an LLM to see all, like, based on the diffs of the code, and then give me a few bullet points on the changes that I made. So it’s easy just, like, for me to copy and paste that into my timesheet.
38 00:09:40.620 ⇒ 00:09:48.990 Pranav Narahari: So, that was just, like, one, like, quick example of, like, using NADN. I’ve messed around it a little bit more as well, and I know, like, there’s a lot of possibilities with it.
39 00:09:48.990 ⇒ 00:09:49.470 Mustafa Raja: Yeah.
40 00:09:49.470 ⇒ 00:10:07.689 Pranav Narahari: But I would say, for the most part, my AI engineering work has been building systems from, like, an idea to then, like, an MVP, and then getting it production ready. So, a lot of the clients that I’ve worked with have asked for exactly that.
41 00:10:08.670 ⇒ 00:10:26.769 Pranav Narahari: there’s been a few instances where I’m, like, joining, like, an existing codebase, but a lot of it is they want me to create an application from scratch that is built on the client’s tech stack. And I think that’s what I feel most comfortable with.
42 00:10:27.090 ⇒ 00:10:34.760 Pranav Narahari: That’s where there’s the most value for clients as well. Not asking them to revamp their entire tooling and start from.
43 00:10:34.760 ⇒ 00:10:35.170 Mustafa Raja: Yeah.
44 00:10:35.170 ⇒ 00:10:45.040 Pranav Narahari: ground zero. It’s, I think clients have really appreciated us just being able to attach to their existing stack and provide value that way.
45 00:10:45.040 ⇒ 00:10:45.850 Mustafa Raja: Yeah.
46 00:10:45.980 ⇒ 00:10:55.540 Mustafa Raja: Yeah, I feel, from scratch… for a few years, it’s going to be, building from scratch, because it’s really a new niche that has came out.
47 00:10:55.750 ⇒ 00:10:57.080 Mustafa Raja: Yeah.
48 00:10:57.190 ⇒ 00:11:22.019 Mustafa Raja: So, what I would want you to, do is explain some sort of a scenario, where you have built something for the client. For example, we have a client. What we do for them is their data structure is spread in way too many sheets and documents, and what we do is we combine, we
49 00:11:22.020 ⇒ 00:11:46.889 Mustafa Raja: there’s some sort of data engineering work involved also. So what we do is we aggregate their data into some sort of Unity, and then we drag it, and then rather than having to look for their data in multiple places, what they now do is they talk to a chatbot and get pretty accurate results over it, and they have reduced a lot of their time that
50 00:11:46.890 ⇒ 00:12:04.140 Mustafa Raja: would be consumed in looking for information around, that has helped the customer service team a lot. So, any sort of automation or AI stuff that you have done, in any sort of niche, I’d love to hear about it.
51 00:12:04.750 ⇒ 00:12:09.149 Pranav Narahari: Totally, something that’s, like, that was reminded
52 00:12:09.330 ⇒ 00:12:15.390 Pranav Narahari: to me, just as you were explaining that, was, a client that I was working with, at a…
53 00:12:15.540 ⇒ 00:12:19.090 Pranav Narahari: just, just recently. Actually, I’m still doing a little bit of work with them.
54 00:12:19.090 ⇒ 00:12:38.159 Pranav Narahari: Is, they’re in the construction space, and they have to process building plans every single day, and they have a dedicated team of, like, 5 annotators whose 9-to-5 job is to go through these building plans and mark the pages that are relevant for their specific business.
55 00:12:38.160 ⇒ 00:12:39.500 Pranav Narahari: And so…
56 00:12:39.940 ⇒ 00:12:50.470 Pranav Narahari: what we did was we built onto their stack, but we used AI to do the processing of finding the relevant pages.
57 00:12:50.470 ⇒ 00:12:50.900 Mustafa Raja: Yeah.
58 00:12:50.900 ⇒ 00:13:00.749 Pranav Narahari: And, what that looked like was their storage… their store of, building plans existed in Microsoft SharePoint.
59 00:13:00.910 ⇒ 00:13:01.280 Mustafa Raja: Yeah.
60 00:13:01.280 ⇒ 00:13:07.300 Pranav Narahari: And so… We created an extension via…
61 00:13:07.560 ⇒ 00:13:23.329 Pranav Narahari: Microsoft SharePoint. They have… I’m not sure if you’ve used SharePoint before, but they have this, like, really niche technology called Power Automate Flows. If you’ve never used SharePoint or built a SharePoint, you probably haven’t heard of it, because I never heard of it before I had… I worked with this client.
62 00:13:23.330 ⇒ 00:13:30.300 Pranav Narahari: But it’s basically a way where you can create custom buttons within, … SharePoint.
63 00:13:30.490 ⇒ 00:13:32.069 Mustafa Raja: Okay. And so….
64 00:13:32.070 ⇒ 00:13:42.670 Pranav Narahari: from a user’s perspective, all they needed to do was to click on this button, and then over the course of some time, because the AI’s gonna take some time to process these sometimes thousand, or…
65 00:13:42.890 ⇒ 00:13:47.200 Pranav Narahari: Upwards of 5,000 page, documents. And then…
66 00:13:47.380 ⇒ 00:14:00.400 Pranav Narahari: they’ll then receive an email when it starts processing, then they’ll receive an email once they’ve… the results of the relevant pages have been uploaded back into SharePoint. So that’s from, like, the customer’s perspective, from…
67 00:14:00.440 ⇒ 00:14:08.910 Pranav Narahari: From a technical design perspective, what was happening was similar to what you were talking about with, finding all of these, like, …
68 00:14:08.980 ⇒ 00:14:09.690 Pranav Narahari: like…
69 00:14:09.690 ⇒ 00:14:34.660 Pranav Narahari: documents in different places. What we did is we built, like, a crawler through their… when I say we, I should just say I was the sole developer on that project. I built a crawler that went into the directory where I know all of the information existed, and based on a few rules that we were able to get from the client, we were able to parse whether, documents were either specification
70 00:14:34.660 ⇒ 00:14:44.370 Pranav Narahari: for a building, or, plans for a building. There’s just, like, two different distinctions for, these, like, building documents.
71 00:14:44.710 ⇒ 00:14:46.139 Pranav Narahari: And, …
72 00:14:46.440 ⇒ 00:14:58.850 Pranav Narahari: they can’t… there’s sometimes just one of each, sometimes it’s hundreds of each. We then… if there was hundreds of each, we would then condense all of these different documents into one long document, and then…
73 00:14:59.360 ⇒ 00:15:06.599 Pranav Narahari: then dispatch that document into an AI engine, and that AI engine would then use OCR for.
74 00:15:06.600 ⇒ 00:15:07.950 Mustafa Raja: Yeah, yeah.
75 00:15:07.950 ⇒ 00:15:20.210 Pranav Narahari: Or plans that are, like, using… that, have a lot of diagrams and extract text that we could then assess to see if it’s relevant for their business.
76 00:15:20.390 ⇒ 00:15:20.990 Pranav Narahari: Yeah.
77 00:15:21.100 ⇒ 00:15:31.980 Mustafa Raja: Yeah, I would be interested in which, … did you use some Langchain to build that AI agent, or what stack were you working with?
78 00:15:32.720 ⇒ 00:15:39.219 Pranav Narahari: Yeah, so, there’s, like, two core com- or three core components to this. One is really small, which is, like, that, …
79 00:15:39.320 ⇒ 00:15:41.220 Pranav Narahari: Yeah, user experience.
80 00:15:41.220 ⇒ 00:15:41.770 Mustafa Raja: She’s just….
81 00:15:41.770 ⇒ 00:15:45.949 Pranav Narahari: button. Yeah, yeah. That was super simple, built that with, …
82 00:15:46.060 ⇒ 00:16:00.520 Pranav Narahari: they have some, like, drag-and-drop thing, it was… but then a little bit of code within SharePoint. Very simple, you’re able to spin that up within, like, less than a day, really. Then there’s the API component to actually dispatch the PDFs and also do some, like.
83 00:16:00.520 ⇒ 00:16:15.460 Pranav Narahari: organization of the PDFs, condensing multiple PDFs into one, and then other types of, like, error handling and guardrails and safety checks. That we built using FastAPI, so fully Python. Yeah.
84 00:16:15.750 ⇒ 00:16:21.860 Pranav Narahari: And then it’s a… it’s, deployed on render. Okay. Okay. Yeah.
85 00:16:21.870 ⇒ 00:16:22.810 Mustafa Raja: And….
86 00:16:22.810 ⇒ 00:16:24.110 Pranav Narahari: Or… yeah.
87 00:16:24.110 ⇒ 00:16:25.239 Mustafa Raja: Yeah, yeah, go on, go on.
88 00:16:25.640 ⇒ 00:16:34.429 Pranav Narahari: Sorry, and then, also for the AI analysis, like, part of the project, that was built using Python as well, …
89 00:16:35.100 ⇒ 00:16:43.889 Pranav Narahari: And, we used… we found that the most accurate was, … OpenAI, for, for this.
90 00:16:43.890 ⇒ 00:16:44.870 Mustafa Raja: Okay, okay, okay.
91 00:16:45.320 ⇒ 00:16:57.850 Mustafa Raja: So, I would be interested in learning if you, if you are looking into, any sort of tools that you are learning for, AI engineering, any new tools that you are looking into.
92 00:16:58.850 ⇒ 00:17:04.149 Pranav Narahari: Yeah, totally. One that’s not too new is, you know, Cursor. I’ve been using.
93 00:17:04.150 ⇒ 00:17:04.549 Mustafa Raja: Oh, yeah.
94 00:17:04.550 ⇒ 00:17:05.110 Pranav Narahari: for the past.
95 00:17:05.119 ⇒ 00:17:12.069 Mustafa Raja: So, I feel, in this age, if a programmer doesn’t know about cursor, they’re not blessed.
96 00:17:12.359 ⇒ 00:17:16.819 Pranav Narahari: Yeah, I totally agree. It really, like, kind of shows you the light, right?
97 00:17:16.819 ⇒ 00:17:17.269 Mustafa Raja: Yo.
98 00:17:17.270 ⇒ 00:17:19.890 Pranav Narahari: Like, it really makes the coding a much more….
99 00:17:19.890 ⇒ 00:17:30.459 Mustafa Raja: I guess, I guess we still need some sort of human touch, because Rayleigh doesn’t have almost all of the context that it needs to have, but it’s getting better with time.
100 00:17:30.780 ⇒ 00:17:34.909 Pranav Narahari: It’s getting better with time, and then I’ve also used Claude Code, I don’t know if you’ve,
101 00:17:34.910 ⇒ 00:17:50.359 Mustafa Raja: Oh, no, I haven’t used, Claude Code. I feel… I really… I really didn’t have to move from cursor. I feel it’s, pretty much doing everything that I would need to do in my use cases, so it’s been pretty simple and pretty good.
102 00:17:50.920 ⇒ 00:18:08.429 Pranav Narahari: Yeah, I would say, I enjoyed using Cloud Code when I wanted to build, user interfaces, because it was a lot quicker to build them. With, Cursor, I was definitely able to build them, and if I compare it to my developer experience from a couple years ago, it’s night and day.
103 00:18:08.430 ⇒ 00:18:28.249 Pranav Narahari: But Claude Code was able to one-shot a lot of dashboards that I wanted to build, much better than Cursor. And Cursor, of course, you can use Claude as the model, however, they throttle the amount of tokens that you’re allowed. With Claude Code, you’re using your own API, so each, inference you’re…
104 00:18:28.250 ⇒ 00:18:33.169 Pranav Narahari: you’re paying a bit of money versus, like, the subscription model that Cursor has. …
105 00:18:33.600 ⇒ 00:18:42.290 Pranav Narahari: And, you know, you can also use your own API key within Cursor as well, but I’ve never had the same experience as I had with Cloud Code.
106 00:18:42.290 ⇒ 00:18:42.800 Mustafa Raja: Yeah.
107 00:18:42.800 ⇒ 00:18:43.350 Pranav Narahari: ….
108 00:18:43.350 ⇒ 00:18:50.199 Mustafa Raja: I’ve heard somewhere that, if we use API with cloud code, it’s going to drain your bank, or something.
109 00:18:50.860 ⇒ 00:18:55.359 Pranav Narahari: A little bit. I wouldn’t say it drains your bank for…
110 00:18:55.600 ⇒ 00:18:59.660 Pranav Narahari: That, dashboard example, it used, I think, $3.
111 00:18:59.850 ⇒ 00:19:01.280 Mustafa Raja: So, yeah, that’s pretty good.
112 00:19:01.280 ⇒ 00:19:01.830 Pranav Narahari: you know.
113 00:19:02.070 ⇒ 00:19:04.149 Pranav Narahari: Pretty good for me, …
114 00:19:04.510 ⇒ 00:19:09.299 Pranav Narahari: At the end of the day, if you use $3 every single day, that’s gonna be almost, what.
115 00:19:09.570 ⇒ 00:19:16.129 Pranav Narahari: 3X or 4X what, … or even more than 4X what cursor is, but…
116 00:19:16.600 ⇒ 00:19:24.860 Pranav Narahari: it’s… I think there’s, certain application for when you want to use it, and, like, you know, even though it’s 4X, like, cursor’s pretty cheap.
117 00:19:25.310 ⇒ 00:19:25.920 Mustafa Raja: Yeah.
118 00:19:26.070 ⇒ 00:19:27.189 Pranav Narahari: Sometimes it’s worth it.
119 00:19:27.230 ⇒ 00:19:32.480 Mustafa Raja: Yeah, so… I feel this is pretty, pretty good, …
120 00:19:32.670 ⇒ 00:19:36.159 Mustafa Raja: I want to know if you know anything about drags or something?
121 00:19:37.590 ⇒ 00:19:38.800 Pranav Narahari: Rag, right?
122 00:19:38.800 ⇒ 00:19:41.349 Mustafa Raja: Yeah, yeah, yeah, retriever Augmented generation.
123 00:19:41.580 ⇒ 00:19:46.309 Pranav Narahari: Exactly, yeah, yeah, yeah. Yeah, I’ve built a few different applications using RAG,
124 00:19:46.390 ⇒ 00:20:03.689 Pranav Narahari: in hackathons, that’s when I first became, like, first learned about it, actually. Or I would say not the first time I learned about it, but the first time I built a project about it. That hackathon was back last September or October,
125 00:20:03.810 ⇒ 00:20:19.970 Pranav Narahari: it was in the Austin area, and the… the goal, or, like, the… the topic that we were competing for was building something that the local Austin population could benefit from. And, I don’t know if you’ve visited Austin before.
126 00:20:20.120 ⇒ 00:20:20.460 Mustafa Raja: Oh, man.
127 00:20:20.460 ⇒ 00:20:23.299 Pranav Narahari: Austin has a… You said you haven’t, or you have?
128 00:20:23.300 ⇒ 00:20:24.509 Mustafa Raja: No, no, no, I haven’t.
129 00:20:24.720 ⇒ 00:20:44.020 Pranav Narahari: Oh, you haven’t? Okay, okay. So, Austin actually has a decent, like, homeless population, and … we built an application that… and we were also looking into this, but a lot of homeless also have phones, and so we thought, okay, why can’t we build an application that maybe uses a local model,
130 00:20:44.440 ⇒ 00:20:47.500 Pranav Narahari: that can… …
131 00:20:47.610 ⇒ 00:20:57.189 Pranav Narahari: Where homeless populations could take a picture of if they have any bruise or rash, and then they can see if it is, like, a really…
132 00:20:57.640 ⇒ 00:21:01.460 Pranav Narahari: If it, if it’s a… what’s the word? If it’s, …
133 00:21:02.220 ⇒ 00:21:07.720 Pranav Narahari: I guess, extremely, important to get diagnosed.
134 00:21:07.720 ⇒ 00:21:08.630 Mustafa Raja: Yeah, yeah.
135 00:21:08.630 ⇒ 00:21:12.929 Pranav Narahari: Or if it’s maybe benign and will go away on its own. …
136 00:21:13.320 ⇒ 00:21:28.910 Pranav Narahari: We… we chose that specifically because we were also provided a few datasets at the hackathon, and we saw this hack… we saw this specific dataset of different, bruises and, like, just, like, skin issues to be, like, an interesting one.
137 00:21:28.910 ⇒ 00:21:35.839 Pranav Narahari: And I was also doing it with one of my non-technical friends that is, in med school, so it seemed like.
138 00:21:35.840 ⇒ 00:21:36.930 Mustafa Raja: Perfect.
139 00:21:36.930 ⇒ 00:21:38.270 Pranav Narahari: Kind of, overlap.
140 00:21:39.090 ⇒ 00:21:41.020 Pranav Narahari: Yeah, and so…
141 00:21:41.280 ⇒ 00:21:50.820 Pranav Narahari: We, we basically used these images, created embeddings, and then did, like, a similarity search, top 5,
142 00:21:51.160 ⇒ 00:21:55.219 Pranav Narahari: And we were able to get pretty accurate results, and …
143 00:21:55.560 ⇒ 00:22:03.470 Pranav Narahari: it was also just, like, a cool experience, because their rag at that point in my life, now I know it’s actually, like, a…
144 00:22:03.740 ⇒ 00:22:06.149 Pranav Narahari: It’s not as, you know….
145 00:22:07.340 ⇒ 00:22:08.010 Mustafa Raja: Mine blockade.
146 00:22:08.010 ⇒ 00:22:08.940 Pranav Narahari: PhD level.
147 00:22:08.940 ⇒ 00:22:10.240 Mustafa Raja: Yeah, yeah, yeah, yeah, I get it.
148 00:22:10.240 ⇒ 00:22:15.999 Pranav Narahari: to know it. It’s actually a pretty simple process, but at that point, it was… it felt really, like, cool to, like.
149 00:22:16.000 ⇒ 00:22:16.420 Mustafa Raja: Yeah.
150 00:22:16.420 ⇒ 00:22:18.029 Pranav Narahari: Build something like that.
151 00:22:18.400 ⇒ 00:22:33.110 Mustafa Raja: Yeah, so for the… you mentioned you ended up using a local model, so what I would be interested in that is, would that model live in the mobile application itself, or would it be accessed through a server?
152 00:22:34.640 ⇒ 00:22:43.730 Pranav Narahari: The local model would have been living on the… on the phone itself. I think we were using, like, the smallest parameter, like, llama model.
153 00:22:43.730 ⇒ 00:22:44.180 Mustafa Raja: Oh.
154 00:22:44.180 ⇒ 00:22:47.229 Pranav Narahari: the hackathon was also, like, sponsored by Llama, so that was.
155 00:22:47.230 ⇒ 00:22:48.969 Mustafa Raja: Oh, yeah, that makes sense.
156 00:22:49.510 ⇒ 00:22:56.609 Pranav Narahari: Yeah. And, we also had it such that, like, if you had internet access, it would use, like, a higher parameter model.
157 00:22:56.920 ⇒ 00:23:08.459 Mustafa Raja: Okay, nice. That’s good. Yeah, I’d be interested in, learning, which database did you use to store the embeddings for RAC?
158 00:23:09.590 ⇒ 00:23:14.830 Pranav Narahari: That’s a good question. There was a sponsor for that…
159 00:23:16.680 ⇒ 00:23:20.870 Pranav Narahari: for that, hackathon, I’m trying to remember, because I haven’t used them since.
160 00:23:21.420 ⇒ 00:23:24.000 Pranav Narahari: Does AstraDB is coming to mind.
161 00:23:24.000 ⇒ 00:23:26.339 Mustafa Raja: Yeah. Have you used… Yeah, yeah, yeah.
162 00:23:26.340 ⇒ 00:23:28.089 Pranav Narahari: I believe it was AstraDB.
163 00:23:28.090 ⇒ 00:23:35.440 Mustafa Raja: Yeah. Yeah, so, we, we most… mostly use, Superbase for most of our stuff, ….
164 00:23:35.440 ⇒ 00:23:36.630 Pranav Narahari: base for all my projects.
165 00:23:36.630 ⇒ 00:23:40.599 Mustafa Raja: Oh, yeah, bro, it’s… it’s super simple and super….
166 00:23:40.600 ⇒ 00:23:41.280 Pranav Narahari: Goodbye.
167 00:23:41.280 ⇒ 00:23:42.110 Mustafa Raja: convenient.
168 00:23:42.400 ⇒ 00:23:43.030 Pranav Narahari: Exactly.
169 00:23:43.030 ⇒ 00:23:46.429 Mustafa Raja: It solved a lot of your pain points and all.
170 00:23:47.000 ⇒ 00:23:47.420 Pranav Narahari: Yeah.
171 00:23:47.420 ⇒ 00:23:50.210 Mustafa Raja: So, let me know if you have any questions for me.
172 00:23:51.070 ⇒ 00:23:52.670 Pranav Narahari: Yeah, …
173 00:23:52.800 ⇒ 00:24:10.860 Pranav Narahari: I’d love to know, like, what the, like, work environment is like. I know that, you know, Brainforge is a relatively new company, you know, I know you guys aren’t too big, but is it, like, pretty collaborative? From what I spoke to with Utham, like, it sounds like it is very collaborative, but, I’m wondering, like, what your experience is so far.
174 00:24:11.050 ⇒ 00:24:30.560 Mustafa Raja: Yeah, so it’s been 3 months, and I’m pretty happy with it. The team that I have is super supportive, so, if I have any sort of question in any way, I’d have instant support from my team members. We have great PMs.
175 00:24:30.560 ⇒ 00:24:37.829 Mustafa Raja: And if you get in, you’d be able to work with them, and you’ll see that, …
176 00:24:38.080 ⇒ 00:24:41.989 Mustafa Raja: PM really, really makes the difference in your life.
177 00:24:41.990 ⇒ 00:24:42.470 Pranav Narahari: I agree.
178 00:24:42.470 ⇒ 00:24:43.810 Mustafa Raja: If, would you…
179 00:24:44.320 ⇒ 00:25:05.279 Mustafa Raja: Let me talk about the clarity on the tickets that I have. I have instant clarity on what I need to do, how it needs to be done, and everything is pretty much sorted, so I just need to look into the ticket, read about it, and it’s clear to me what needs to be done, where it needs to be done, how it needs to be done.
180 00:25:05.280 ⇒ 00:25:26.290 Mustafa Raja: And even if after that I have a question, it’s, pretty quickly answered. So, and the team environment is really, really friendly. I’d say, and I did, say to many of my team members that this, really, is the dream environment that I wanted.
181 00:25:26.290 ⇒ 00:25:26.710 Pranav Narahari: Wow.
182 00:25:26.710 ⇒ 00:25:31.329 Mustafa Raja: … It’s just that they’re… they’re really friendly.
183 00:25:31.800 ⇒ 00:25:32.680 Pranav Narahari: That’s awesome.
184 00:25:32.680 ⇒ 00:25:34.769 Mustafa Raja: Yeah. Yeah, any other questions?
185 00:25:35.480 ⇒ 00:25:42.770 Pranav Narahari: Yeah, I guess, so I saw that there’s a few roles on the Brainforge website. What would you say would be, like.
186 00:25:43.750 ⇒ 00:25:48.520 Pranav Narahari: is the need right now. I know it’s an early company, so there’s probably, like.
187 00:25:48.670 ⇒ 00:26:04.089 Pranav Narahari: probably a lot of needs, people are probably wearing a lot of hats, but if, you know, an AI engineer were to, like, come in, what do you think they would be working on? And also, what do you think would be, like, an asset in terms of skill set?
188 00:26:05.040 ⇒ 00:26:12.130 Mustafa Raja: Yeah, so, we do have, an opening for AI Engineer, and…
189 00:26:12.160 ⇒ 00:26:27.670 Mustafa Raja: what I sort of started on was the internal platform only. I worked on internal tools, built a few of them for sales, for PMs, and for operations, and then slowly started moving on client work.
190 00:26:27.670 ⇒ 00:26:37.469 Mustafa Raja: And I think this… this is going to be… if you were to join us, this is going to be a pretty similar path for you. You’ll start working on our internal stuff.
191 00:26:37.470 ⇒ 00:26:41.820 Mustafa Raja: And then slowly, move towards the clients.
192 00:26:42.780 ⇒ 00:26:46.790 Pranav Narahari: That’s awesome. Okay. Yeah, that seems like a great progression, just so, like.
193 00:26:46.790 ⇒ 00:26:47.150 Mustafa Raja: Right, yeah.
194 00:26:47.390 ⇒ 00:26:48.170 Pranav Narahari: It seems like….
195 00:26:48.170 ⇒ 00:27:07.329 Mustafa Raja: Just to understand the tools. Yeah, just to understand the tools, what we work with, because, for whatever we are doing for the clients, it’s pretty much the same stack we use internally. So in terms of stack, you wouldn’t have any issues learning new, learning new tools, or anything like that.
196 00:27:07.650 ⇒ 00:27:08.340 Pranav Narahari: Yeah.
197 00:27:08.390 ⇒ 00:27:13.470 Mustafa Raja: Yeah, we… for AI engineering, we mostly use NHN.
198 00:27:13.470 ⇒ 00:27:13.810 Pranav Narahari: Okay.
199 00:27:13.810 ⇒ 00:27:25.120 Mustafa Raja: database, you already know that we use, Superbase, and then we have AWS and Snowflake for some other things, and, for some scheduling, we use Dagster.
200 00:27:25.540 ⇒ 00:27:33.019 Mustafa Raja: Daxa is actually a job scheduler, for Python scripts.
201 00:27:33.580 ⇒ 00:27:34.390 Pranav Narahari: I see.
202 00:27:34.390 ⇒ 00:27:37.080 Mustafa Raja: Yeah, let me know if you have any other questions.
203 00:27:37.950 ⇒ 00:27:44.939 Pranav Narahari: I think that’s most of the questions I have right now, … Would you say that, …
204 00:27:45.140 ⇒ 00:27:51.329 Pranav Narahari: Actually, one question, like, a really basic question is, how many AI engineers are there at, Brainforge currently?
205 00:27:51.900 ⇒ 00:27:54.409 Mustafa Raja: Three. Three.
206 00:27:54.410 ⇒ 00:27:54.840 Pranav Narahari: Okay, good.
207 00:27:54.840 ⇒ 00:28:00.950 Mustafa Raja: So, one of, one of the ones, left, this month.
208 00:28:02.100 ⇒ 00:28:06.720 Mustafa Raja: So, there’s currently 3 in… there’s an opening for one.
209 00:28:07.490 ⇒ 00:28:10.899 Pranav Narahari: Gotcha. Okay, that’s really cool. And …
210 00:28:12.620 ⇒ 00:28:20.319 Pranav Narahari: is the… you said, like, it’s pretty collaborative, and based on what I talked to Itham about was that it was collaborative. That’s kind of what I’m looking for, because.
211 00:28:20.320 ⇒ 00:28:20.930 Mustafa Raja: Again, my….
212 00:28:20.930 ⇒ 00:28:24.760 Pranav Narahari: current roles, it’s not very collaborative, and …
213 00:28:24.930 ⇒ 00:28:30.810 Pranav Narahari: It makes sense, also, like, why it’s difficult, because when you’re working with a client, like.
214 00:28:31.010 ⇒ 00:28:36.119 Pranav Narahari: there’s some time… it’s difficult to then talk to people on different clients, right? Because, you know, time is finite.
215 00:28:36.120 ⇒ 00:28:36.750 Mustafa Raja: Yeah.
216 00:28:36.750 ⇒ 00:28:41.660 Pranav Narahari: So, how would you say that, Brainforge does that?
217 00:28:42.500 ⇒ 00:28:47.459 Mustafa Raja: So, so far, we…
218 00:28:47.460 ⇒ 00:29:05.459 Mustafa Raja: We have, three PMs. One of them is Utam, and, there’s one other, Amber that I work with, and, Utam sort of manages, all of the tickets, that, I would have for the clients that he’s managing.
219 00:29:05.460 ⇒ 00:29:19.990 Mustafa Raja: And Amber does for the other ones, and we also… for one client, we, there’s one of the AI engineers that is working with me, and he has, most of the contacts about
220 00:29:19.990 ⇒ 00:29:37.009 Mustafa Raja: about almost everything, so I’ll just go and message him if I am confused, or I need to look for something that, I don’t know is there. I’ll just message him, and he’ll guide me what to look for, where to look for.
221 00:29:37.660 ⇒ 00:29:40.100 Mustafa Raja: And that sort of stuff. So, …
222 00:29:40.230 ⇒ 00:29:47.189 Mustafa Raja: The crux of this is the team is very responsive in answering your questions, to unblock you if you are blocked on anything.
223 00:29:48.530 ⇒ 00:29:50.739 Pranav Narahari: That’s super important to me, so that’s awesome.
224 00:29:50.940 ⇒ 00:30:03.860 Mustafa Raja: Yeah. So, the goal of the team is people aren’t blocked on something that is pretty simple or can be answered in a few minutes. They’ll reach out to you, help you with whatever you need.
225 00:30:04.890 ⇒ 00:30:12.840 Pranav Narahari: Gotcha. Okay, that’s, I mean, that’s, honestly more than you can ask for at most companies, so that’s awesome.
226 00:30:13.290 ⇒ 00:30:17.109 Mustafa Raja: Yeah, yeah, and we currently work remotely.
227 00:30:17.620 ⇒ 00:30:18.230 Pranav Narahari: Yep.
228 00:30:18.500 ⇒ 00:30:20.570 Mustafa Raja: I’m based in Pakistan.
229 00:30:20.880 ⇒ 00:30:30.160 Pranav Narahari: Oh, okay. Yeah, I noticed on the actual, like, the… on the invite, it said, like, 11.30 PM, and then, like, Pakistan Standard Time, so….
230 00:30:30.160 ⇒ 00:30:30.530 Mustafa Raja: Yeah.
231 00:30:30.830 ⇒ 00:30:33.099 Pranav Narahari: Yeah, yeah, so it’s pretty late for you then.
232 00:30:33.900 ⇒ 00:30:41.640 Mustafa Raja: Yeah, I was actually… I’m actually a night out, and this worked out pretty good for me. This opportunity really worked out pretty good for me.
233 00:30:41.890 ⇒ 00:30:42.730 Pranav Narahari: Yeah.
234 00:30:42.860 ⇒ 00:30:45.189 Pranav Narahari: How did you, how did you find Brainforge?
235 00:30:45.190 ⇒ 00:31:03.720 Mustafa Raja: Oh, yeah, that’s a pretty good story. So, I just graduated this, this year, and, while I was in, in my grad studies, what I did, I used to freelance for most, most of my, teachers.
236 00:31:03.800 ⇒ 00:31:21.319 Mustafa Raja: They would have some sort of clients and pitch me to them, and then I’ll get the client and work on that. I started from Web React, moved to React Native for mobile applications, and then when AI engineering started taking Bloom.
237 00:31:21.320 ⇒ 00:31:30.939 Mustafa Raja: YouTube would start recommending me langchain videos, and that’s sort of how I got to know about this particular field. Yeah.
238 00:31:30.940 ⇒ 00:31:45.340 Mustafa Raja: looked into a few videos, loved it, decided I’m going to move to this, and started looking for clients. Worked for a few clients, and then that is how, Utem found me on Discord.
239 00:31:46.140 ⇒ 00:31:50.030 Pranav Narahari: Oh, oh, he told me the story, actually, is that he found somebody on Discord.
240 00:31:50.030 ⇒ 00:31:50.510 Mustafa Raja: That’s awesome.
241 00:31:50.510 ⇒ 00:31:52.850 Pranav Narahari: Awesome. That’s really cool. Yeah.
242 00:31:52.850 ⇒ 00:32:01.959 Mustafa Raja: So, I was actually client hunting on Discord, stumbled into Utham, and we talked, and a few months later, I was in DreamForge.
243 00:32:02.690 ⇒ 00:32:06.080 Pranav Narahari: Oh, wow. That’s a really funny story. I’ve never heard of somebody…
244 00:32:06.250 ⇒ 00:32:08.440 Pranav Narahari: Getting hired off Discord. That’s really cool.
245 00:32:08.440 ⇒ 00:32:13.429 Mustafa Raja: Yeah, I haven’t… I haven’t… I didn’t even have it in my mind that I could be….
246 00:32:13.790 ⇒ 00:32:16.229 Pranav Narahari: Yeah. … Yeah.
247 00:32:16.600 ⇒ 00:32:20.230 Mustafa Raja: Yeah, this is pretty good, this is pretty good. Yeah, yeah, do you have any questions?
248 00:32:21.110 ⇒ 00:32:33.129 Pranav Narahari: Yeah, just one more question. With, AI engineering, like, at, Brainforge, is it… are most of them, like, part-time? Putnam mentioned how it would start, like, part-time, and then.
249 00:32:33.130 ⇒ 00:32:33.700 Mustafa Raja: Yeah, yeah, yeah.
250 00:32:33.700 ⇒ 00:32:34.070 Pranav Narahari: if I were to.
251 00:32:34.070 ⇒ 00:32:34.440 Mustafa Raja: So….
252 00:32:34.440 ⇒ 00:32:35.120 Pranav Narahari: role.
253 00:32:35.120 ⇒ 00:32:53.629 Mustafa Raja: Yeah, yeah, so how I started. I started at 10 hours, per week, and within 2 weeks, I was bumped to 20, and then 30, and then 40. So 40 is a full-time, right? So, I was, bumped to 40 within a month or so.
254 00:32:53.810 ⇒ 00:32:54.780 Mustafa Raja: Okay.
255 00:32:54.780 ⇒ 00:32:55.330 Pranav Narahari: So, a week over.
256 00:32:55.330 ⇒ 00:33:00.999 Mustafa Raja: Yeah, yeah, week over week. I feel it’s going to be pretty similar for you, if that would be the case.
257 00:33:02.310 ⇒ 00:33:04.069 Pranav Narahari: I don’t know if you have any other questions.
258 00:33:04.670 ⇒ 00:33:10.019 Pranav Narahari: I think that’s all I have right now. If anything comes up, I’ll just reach out via email, does that work?
259 00:33:10.020 ⇒ 00:33:16.729 Mustafa Raja: Yeah, yeah, that works really good. And this was the first interview, right?
260 00:33:17.470 ⇒ 00:33:29.029 Pranav Narahari: Right, yeah, so I had, like, an initial call with them, like, on Friday of last week, and then this is the first one of this week, and then I have a call with Amber on Friday, and I believe that’s the last one.
261 00:33:29.030 ⇒ 00:33:30.000 Mustafa Raja: Yeah, yeah.
262 00:33:30.290 ⇒ 00:33:30.900 Pranav Narahari: Yeah.
263 00:33:31.160 ⇒ 00:33:41.879 Mustafa Raja: Okay, this is pretty good. One last thing, I would want to ask you is, are you familiar with Langchain or such tools?
264 00:33:42.690 ⇒ 00:33:43.330 Pranav Narahari: Yes, I don’.
265 00:33:43.330 ⇒ 00:33:46.359 Mustafa Raja: We don’t use it, but, just…
266 00:33:46.700 ⇒ 00:33:50.270 Mustafa Raja: it has a lot of context about AI engineering.
267 00:33:51.040 ⇒ 00:33:59.669 Pranav Narahari: Yeah, so I’ve used a link chain for when I was building, like, another, like, RAG project, …
268 00:33:59.780 ⇒ 00:34:04.710 Pranav Narahari: I have to look back into that project, like, how I was using it, but…
269 00:34:04.890 ⇒ 00:34:10.230 Pranav Narahari: I remember I was using, like, some, like, algorithm for…
270 00:34:10.360 ⇒ 00:34:16.730 Pranav Narahari: embeddings. That was, like, a super basic algorithm, so I wanted to, like, just understand, like, the bare bones of, like, what is… what is, like.
271 00:34:16.730 ⇒ 00:34:17.060 Mustafa Raja: Yeah.
272 00:34:17.060 ⇒ 00:34:20.810 Pranav Narahari: headings even mean. It was, like, some, sentence transformer.
273 00:34:21.040 ⇒ 00:34:22.050 Mustafa Raja: Oh, yeah, yeah, yeah.
274 00:34:22.050 ⇒ 00:34:27.379 Pranav Narahari: Yeah, so, like, super basic stuff, but I wanted to not use, like, a ton of third-party libraries, I wanted to just, like….
275 00:34:27.389 ⇒ 00:34:29.809 Mustafa Raja: Okay, so you wanted to build your own thing.
276 00:34:30.090 ⇒ 00:34:34.929 Pranav Narahari: Yeah, and so that’s kind of my experience with Langchain, like, just, like, really basic….
277 00:34:34.929 ⇒ 00:34:40.650 Mustafa Raja: Yeah, we don’t use lines in over here, it takes a lot of time to set stuff, and….
278 00:34:41.090 ⇒ 00:34:41.500 Pranav Narahari: Yeah.
279 00:34:42.090 ⇒ 00:34:55.729 Mustafa Raja: So, and then it really works good in that space. Pretty, pretty easy and pretty quick to set up, so the deliverables are, are in reasonable time also.
280 00:34:55.860 ⇒ 00:34:58.980 Mustafa Raja: Yeah, this is pretty good. Thank you so much. Thank you so much for joining us.
281 00:34:58.980 ⇒ 00:35:01.700 Pranav Narahari: Yeah, I’ll talk to you soon, hopefully.
282 00:35:01.820 ⇒ 00:35:04.399 Mustafa Raja: Yeah, hopefully. Thank you so much, have a good day.
283 00:35:04.750 ⇒ 00:35:05.839 Pranav Narahari: Yeah, you too.
284 00:35:05.840 ⇒ 00:35:07.109 Mustafa Raja: Thank you, bye-bye.
285 00:35:07.110 ⇒ 00:35:07.680 Pranav Narahari: Bye.