Meeting Title: Brainforge Interview w- Sam Date: 2026-03-04 Meeting participants: Srinivas Saiteja Tenneti’s Calendly Notetaker, Srinivas Saiteja Tenneti, Samuel Roberts
WEBVTT
1 00:00:13.200 ⇒ 00:00:14.090 Samuel Roberts: Hello?
2 00:00:14.420 ⇒ 00:00:16.180 Srinivas Saiteja Tenneti: Hello, how are you doing?
3 00:00:16.440 ⇒ 00:00:17.899 Samuel Roberts: Good, good. How about yourself?
4 00:00:18.230 ⇒ 00:00:21.619 Srinivas Saiteja Tenneti: I’m doing great. Can you see my, like, am I audible clearly?
5 00:00:21.860 ⇒ 00:00:25.239 Samuel Roberts: Yes, I can hear you, I can see you. Am I good as well?
6 00:00:25.540 ⇒ 00:00:29.890 Srinivas Saiteja Tenneti: Yeah, yeah, I’m just checking, like, if anything is… everything is perfect, yeah.
7 00:00:29.890 ⇒ 00:00:36.699 Samuel Roberts: No, you’re good, you’re good. I always have technical issues on these calls. That’s why I have a new mic, because my old one wasn’t working all the time, so…
8 00:00:36.700 ⇒ 00:00:45.970 Srinivas Saiteja Tenneti: Okay, there is something, like, give me a second. Calendly is joined, I guess. I don’t know why, I… I’m using, like, Calendly to schedule meetings.
9 00:00:46.310 ⇒ 00:00:48.649 Samuel Roberts: Oh, okay, yeah, it looks like it has a note-taker here.
10 00:00:48.650 ⇒ 00:00:51.470 Srinivas Saiteja Tenneti: So, should I stop it, or is it fine, like,
11 00:00:51.890 ⇒ 00:00:53.680 Srinivas Saiteja Tenneti: How to, like, stop it, I don’t know.
12 00:00:54.600 ⇒ 00:00:56.339 Samuel Roberts: That I don’t know.
13 00:00:56.820 ⇒ 00:01:00.109 Srinivas Saiteja Tenneti: I don’t know how Calendly came here. Give me a second.
14 00:01:00.110 ⇒ 00:01:01.290 Samuel Roberts: Okay, sure.
15 00:01:02.340 ⇒ 00:01:04.579 Srinivas Saiteja Tenneti: If you want, I will try to stop it.
16 00:01:05.269 ⇒ 00:01:23.299 Samuel Roberts: If you… if you need to, if you want to, it’s… if you… don’t… don’t spend too much time. If you need to, it’s fine, but if you… if it doesn’t work, we can leave it. You’re good. It is just like a meeting, so, you know, schedule meetings via Calendly, so yeah. Yeah, yeah, okay. Anyway, let’s, let’s jump in. So, nice to meet you. My name is…
17 00:01:23.460 ⇒ 00:01:33.900 Samuel Roberts: Sam Roberts, I’m the AI tech lead here at Brainforge. I think we’ll just start with an introduction from you, and then I have a bunch of questions, I’ll make some time for you to ask questions, and yeah.
18 00:01:33.900 ⇒ 00:01:34.540 Srinivas Saiteja Tenneti: Costa.
19 00:01:34.690 ⇒ 00:01:41.349 Srinivas Saiteja Tenneti: Hi Sam, my name is Srinvas Saitejat Nati. I know it’s long to spell, so you can call me Srini.
20 00:01:41.360 ⇒ 00:01:53.609 Srinivas Saiteja Tenneti: So, I’m an AIML engineer right now, with about, like, 4 years of experience working, like, machine learning techniques, and also, like, as a data scientist too, so… and also, like, proper Gen AI techniques also.
21 00:01:53.610 ⇒ 00:02:02.819 Srinivas Saiteja Tenneti: So, right now, I work at UnitedHealthcare, where I build, like, AI applications using, like, LLMs, like GPT-4, the project, right?
22 00:02:02.820 ⇒ 00:02:12.449 Srinivas Saiteja Tenneti: So, one of the main projects I worked on was a healthcare AI assistant that helps, like, our doctors and clinical staff quickly, like, find information from
23 00:02:12.450 ⇒ 00:02:26.190 Srinivas Saiteja Tenneti: medical guidelines and patient policies. It’s not only, like, an, you know, internal healthcare AI assistant, but also, like, it is useful for, like, later on, we upgraded it for the new joiners also, like, new boarders, to check the relevant docs.
24 00:02:26.190 ⇒ 00:02:46.059 Srinivas Saiteja Tenneti: usually, like, what happens is that when new joiners, or onboarders, or let’s say the working people want to look through the share, like, documents, they go to SharePoint, or reliant databases, it takes time. So instead of that, the chatbot, like, minimizes the time it usually takes to search the docs and what all the docs are there.
25 00:02:46.060 ⇒ 00:02:52.949 Srinivas Saiteja Tenneti: So it is like an… it simply, like, you know, reduces the time, and also it helps to give the relevant documents.
26 00:02:52.950 ⇒ 00:03:09.919 Srinivas Saiteja Tenneti: And it also gives the medical guidelines and also the patient policies and relevant clinical documents also. I built this system using, like, a proper RAG architecture, using, like, LangChain, and also vector search, so the model can retrieve the right information before answering questions.
27 00:03:09.990 ⇒ 00:03:18.560 Samuel Roberts: Before that, I worked as a data scientist at Science, which I have told, like, there I built machine learning models, like, including a customer churn prediction system.
28 00:03:18.560 ⇒ 00:03:23.149 Srinivas Saiteja Tenneti: That helped the business identify customers who might, like, you know, leave
29 00:03:23.220 ⇒ 00:03:39.970 Srinivas Saiteja Tenneti: Or else, like, who might… who are about to leave. So, this helped, like, the company take action early. I also worked on data preprocessing and data processing techniques, and also, like, model development, deployment, and cloud systems using, like, Python and AWS, which are my…
30 00:03:39.970 ⇒ 00:03:46.210 Srinivas Saiteja Tenneti: core, I would say. I have used it the most, but of course, I have hands-on on Azure and other systems also, so…
31 00:03:46.220 ⇒ 00:03:51.990 Samuel Roberts: I even Google, like, those kind of, like, Google Garden, and these, I have, like, model garden.
32 00:03:51.990 ⇒ 00:03:53.729 Srinivas Saiteja Tenneti: I’ve used a lot of tools.
33 00:03:53.730 ⇒ 00:03:55.960 Samuel Roberts: I do, like, personal projects also.
34 00:03:55.960 ⇒ 00:04:09.260 Srinivas Saiteja Tenneti: Overall, I enjoy building, like, you know, end-to-end AI solutions, especially, like, using LLMs. Sometimes, you know, like, if I see some link in context, I would, like, search for SLMs and, like, you know, brainstorm and those things.
35 00:04:09.260 ⇒ 00:04:09.770 Samuel Roberts: Sure.
36 00:04:09.770 ⇒ 00:04:17.830 Srinivas Saiteja Tenneti: Yeah, and machine learning, and of course, cloud technologies. So, these I’m very good at, and also, like, I’m good in AI automation also, so yeah.
37 00:04:18.320 ⇒ 00:04:18.910 Samuel Roberts: Great.
38 00:04:19.240 ⇒ 00:04:21.909 Samuel Roberts: So let’s, let’s talk about,
39 00:04:22.220 ⇒ 00:04:32.279 Samuel Roberts: what part of the stack do you… or the AI stack in general, because stack could mean a few different things here, but for AI tooling, what have you spent the most time building on versus just experimenting with?
40 00:04:32.740 ⇒ 00:04:48.960 Srinivas Saiteja Tenneti: I mean, if you ask me particularly, like, what stack, I have used it, like, in an experiment… experience perspective. I would say, like, I have used it Fais, Chroma also, if you, like, like, and also, like, Pinecone. Recently, like, I’m using Pinecone for the vector store.
41 00:04:49.060 ⇒ 00:05:01.960 Srinivas Saiteja Tenneti: And, like, for the… this, bot which I have built, for that, I have used it files as a vector store. Apart from that, for the, like, Evidently AI for AI evaluation monitoring, I have used it Evidently AI.
42 00:05:02.080 ⇒ 00:05:07.209 Srinivas Saiteja Tenneti: And, the tools like, you know, basic, like, AWS services, which we use.
43 00:05:07.230 ⇒ 00:05:19.919 Srinivas Saiteja Tenneti: And monitoring, of course, I use Datadog because you can easily build dashboards. Of course, we have CloudWatch, but CloudWatch, we can get, like, PE95, those response times, and logs also in CloudWatch.
44 00:05:19.920 ⇒ 00:05:32.390 Srinivas Saiteja Tenneti: And Datadog, of course, we get logs, but also we can clearly, like, you know, build the… I mean, dashboards. From my personal perspective, I have used it, like, for my… right now, I’m building a few interesting projects.
45 00:05:32.410 ⇒ 00:05:43.299 Srinivas Saiteja Tenneti: For that, I’m using SupaBase, and also, like, a few tools, like, Cohere and, Elevenlabs. So these are, like, few tools which I’m merging and trying to do something, like.
46 00:05:43.300 ⇒ 00:05:43.710 Samuel Roberts: Okay.
47 00:05:43.710 ⇒ 00:05:58.870 Srinivas Saiteja Tenneti: for an athletic performance, let’s say any person uploads their gym or athletic videos or anything, it will, like, give you… it will analyze it and give you a proper end-to-end plan, like a weekly, monthly plan, so… I’m trying to build that, so, yeah.
48 00:05:58.870 ⇒ 00:05:59.520 Samuel Roberts: Nice.
49 00:05:59.650 ⇒ 00:06:00.500 Samuel Roberts: Great.
50 00:06:00.990 ⇒ 00:06:18.429 Samuel Roberts: Okay, let’s take a step back a little bit and talk about, you know, this stuff’s moving so quickly, the tech’s changing all the time, so a lot of non-technical people hear all kinds of things. So how do you explain the limitations of an LLM or a tool to a non-technical stakeholder?
51 00:06:19.030 ⇒ 00:06:32.109 Srinivas Saiteja Tenneti: Because whenever, like, you know, it’s not about, like, not only non-technical, let’s say I, like, say, from my perspective, when I started to getting new about AI, no, for me, it looked like, you know, huge and something, like, weird.
52 00:06:32.110 ⇒ 00:06:32.720 Samuel Roberts: Sure.
53 00:06:32.720 ⇒ 00:06:39.409 Srinivas Saiteja Tenneti: But after learning, it became, like, clear to me. But as to a, let’s say, a non-technical stakeholder is there.
54 00:06:39.440 ⇒ 00:06:52.059 Srinivas Saiteja Tenneti: So, for them, I try to explain in simple and practical terms, so non-technical stakeholders can easily understand it. I normally tell when LLM is very good at understanding language.
55 00:06:52.060 ⇒ 00:07:02.419 Srinivas Saiteja Tenneti: and generating answers, okay? So this is what I would say, but it is not the same as a system that truly understands facts, like how we do, like how you humans do.
56 00:07:02.420 ⇒ 00:07:15.969 Srinivas Saiteja Tenneti: Right. So, it is, like, something, like, okay, it usually, like, okay, understand language. It is, like, it, of course, it, works on NLP techniques, like, I mean, it answers in, like, proper, like, human language.
57 00:07:15.970 ⇒ 00:07:35.010 Srinivas Saiteja Tenneti: So, the model learns patterns from a lot of text data, so sometimes it can give very confident answers, and that are not always correct, of course. So, this is what we call as hallucination, and where the model generates information that sounds right, you know, may… but may not be accurate.
58 00:07:35.060 ⇒ 00:07:44.960 Srinivas Saiteja Tenneti: So, this is what I would say, and I also explained that LLMs work, best when they have, like, good context, or data, or a tool like Refer.
59 00:07:45.150 ⇒ 00:07:55.819 Srinivas Saiteja Tenneti: And that’s why in many projects we use, like, approaches like RAG, which is, like, a retrieval augmented generation, where the system retrieves information from trusted documents or, like.
60 00:07:55.940 ⇒ 00:08:15.900 Srinivas Saiteja Tenneti: I mean, or databases, before the model generates a response. So, like, if you tell me, like, if you ask me, like, how are you going to tell your dad, who don’t know about technology or anything, I would simply tell, like, if I am, it is something like my clone, okay? But not completely as a clone.
61 00:08:16.020 ⇒ 00:08:20.579 Srinivas Saiteja Tenneti: It just, okay, it do whatever I tell it. It is, like, trained to it.
62 00:08:20.630 ⇒ 00:08:36.969 Srinivas Saiteja Tenneti: Okay, but whatever it gives it, it’s not, like, completely true or, like, like, you know? It is, like, it behaves well, it should be trained well, it should be, like, how you, like, create a clone and train it well. So, in that way, it is. But, still, it can’t beat our human brain, I would say.
63 00:08:37.599 ⇒ 00:08:39.799 Samuel Roberts: Sure. Alright, great.
64 00:08:40.379 ⇒ 00:08:46.709 Samuel Roberts: Okay, I’m trying to think where I want to go next. There’s a few different places we can go, but let’s… let’s talk about,
65 00:08:47.119 ⇒ 00:08:50.909 Samuel Roberts: So say you’ve built something for a stakeholder like this, and it,
66 00:08:51.049 ⇒ 00:08:59.029 Samuel Roberts: Maybe they misunderstood what it could do? Okay. Is there… do you have any times that that’s happened that you can tell me about?
67 00:08:59.630 ⇒ 00:09:05.260 Srinivas Saiteja Tenneti: Yeah, of course. So, basically, one instance, what happened is that, you know, for the medical chatbot only.
68 00:09:05.700 ⇒ 00:09:22.389 Srinivas Saiteja Tenneti: So, one of the stakeholders, I don’t, like, he was, like, you know, he, he was, like, learning about, he was, I mean, you know, reading about newsletters, we say AI newsletters and everything. So, at the time, he suggested the top, you know, like, LLMs, or top tools.
69 00:09:22.390 ⇒ 00:09:29.819 Srinivas Saiteja Tenneti: which were very famous at that time, so he was suggesting that you should use that, it might be helpful. So these things happened, so…
70 00:09:29.820 ⇒ 00:09:30.390 Samuel Roberts: Yeah, yeah.
71 00:09:30.390 ⇒ 00:09:41.290 Srinivas Saiteja Tenneti: If a stakeholder knows these kind of tools, they usually suggest, like, why aren’t you using this? I have learned in the recent newsletter that this is really great, this works that.
72 00:09:41.290 ⇒ 00:09:52.980 Srinivas Saiteja Tenneti: But what I have, like, told them that, see, in some perspective, if we use a lot of tools, it might be, you know, like, create confusion. So for a simple pipeline, for a simple architecture.
73 00:09:52.980 ⇒ 00:10:06.609 Srinivas Saiteja Tenneti: The basic ones with proper, you know, if it is reliable, and if it is, like, giving the proper retrieval, and if it has a proper retrieval scores, and if it is giving proper answers, and everything.
74 00:10:06.610 ⇒ 00:10:18.440 Srinivas Saiteja Tenneti: Then, of course, the smaller one works well. But, if we, like, sometimes if, you know, if there is a scope, of course, we will be, like, introducing, but if there is no… not required for the scope, if we, like, you know…
75 00:10:18.610 ⇒ 00:10:35.249 Srinivas Saiteja Tenneti: if we, like, put it a lot, then of course, it will be… become messy. And one, like, one more example, like, when we were building that healthcare AI assistant only for clinical staff, some… also few stakeholders, like, initially thought the system would behave like, you know, like a fully intelligent medical expert.
76 00:10:35.340 ⇒ 00:10:44.369 Srinivas Saiteja Tenneti: That could answer any healthcare question perfectly, but in reality, the system was designed to, you know, retrieve information from specific
77 00:10:44.390 ⇒ 00:10:56.969 Srinivas Saiteja Tenneti: like, documents or medical guidelines and internal docs, not to, like, replace a doctor or provide medical conditions. In healthcare, this is what I told to every stakeholder, that we need to follow HIPAA rules.
78 00:10:57.040 ⇒ 00:11:11.510 Srinivas Saiteja Tenneti: And, the bot shouldn’t, like, replace doctors completely, because if it replaces, sometimes it can mislead, and it will cause a great problem. So, this is what I have told in them in a simple way. So, yeah.
79 00:11:11.510 ⇒ 00:11:14.290 Samuel Roberts: Great. These things happen, actually. Oh, yeah.
80 00:11:14.400 ⇒ 00:11:15.060 Srinivas Saiteja Tenneti: Yeah.
81 00:11:15.060 ⇒ 00:11:19.340 Samuel Roberts: Oh, yeah. Okay, let’s see,
82 00:11:19.790 ⇒ 00:11:22.659 Samuel Roberts: Alright, let’s talk about… I’ve mentioned it before, but, like, things are…
83 00:11:22.880 ⇒ 00:11:24.489 Srinivas Saiteja Tenneti: Yeah, of course.
84 00:11:24.610 ⇒ 00:11:28.190 Srinivas Saiteja Tenneti: Sorry, hello? Hello? Your, your voice muted.
85 00:11:28.190 ⇒ 00:11:42.300 Samuel Roberts: There’s a very… my mic is very sensitive when I touch it, it mutes itself. Can you talk to me about a trend, or a tool or framework that you were excited about, but decided not to adopt for a project for any reason?
86 00:11:42.680 ⇒ 00:11:49.929 Srinivas Saiteja Tenneti: I mean, see, like, trend or tool, if you… there are a lot of, like, you know, anti-gravity, which came, like, in January.
87 00:11:49.930 ⇒ 00:11:50.630 Samuel Roberts: Sure.
88 00:11:50.630 ⇒ 00:12:01.770 Srinivas Saiteja Tenneti: And, I mean, not Jan, it came in November, but slowly, like, introduced by, like, Jan and everyone are using. And also, like, there are a few, like, if you ask me…
89 00:12:01.810 ⇒ 00:12:16.360 Srinivas Saiteja Tenneti: I wanted, you know, whatever the tools or new tools are there, I don’t want to directly, like, you know, use it in the enterprise way, like, enterprise sector. I just want to try it by using, like, you know, in my own personal projects.
90 00:12:16.360 ⇒ 00:12:24.459 Srinivas Saiteja Tenneti: Then, if I’m sure about it, if it is working well, if it could replace something which I previously built in an enterprise setting, then I will be like.
91 00:12:24.460 ⇒ 00:12:28.670 Srinivas Saiteja Tenneti: having a short meeting, and I will discuss with them, and then I will take it on.
92 00:12:28.740 ⇒ 00:12:45.400 Srinivas Saiteja Tenneti: But usually, like, I’m… right now, I’m exploring, like, Cohere, and also, like, 11 Labs, which is… and also, like, Supabase, and as I’ve told, and… and also, like, Pinecone is also great, which is not, latest one, but it is, of course, there from long.
93 00:12:45.540 ⇒ 00:12:57.980 Srinivas Saiteja Tenneti: And, that actually explains, you know, like, that actually happens quite often, because AI ecosystem is changing very fast. Not about, like, you know, monthly, monthly, it is, like, daily, daily, hourly.
94 00:12:58.810 ⇒ 00:13:09.310 Srinivas Saiteja Tenneti: And not every tool is the right fit for a real production project. So, one example, if you ask, like, another example, I could say, like, AI agents frameworks for a project, like.
95 00:13:09.310 ⇒ 00:13:20.450 Srinivas Saiteja Tenneti: when we were exploring, at that time, tools that allow LLMs to act as autonomous agents were becoming very popular. The idea was, like, interesting because agents can plan
96 00:13:20.450 ⇒ 00:13:34.069 Srinivas Saiteja Tenneti: like, steps, like, you know, call tools and automate workflows. However, when we evaluated it for our project, we realized that the behavior of the agents was not always predictable. This happened when we were building the chatbot.
97 00:13:34.300 ⇒ 00:13:50.499 Srinivas Saiteja Tenneti: So, at that time, like, it was new, like, agentic AI was new, right? So sometimes the model would take unnecessary, like, steps or produce inconsistent results, so we can’t entertain those scenarios when we are building, like, in a secure environment.
98 00:13:50.500 ⇒ 00:14:01.039 Srinivas Saiteja Tenneti: So, since the project involved healthcare information, we needed the system to be, like, you know, pretty reliable, and I would say controlled, because it should follow HIPAA rules also.
99 00:14:01.430 ⇒ 00:14:20.469 Srinivas Saiteja Tenneti: And, so instead of using a full agent framework, we decided to build a more structured RAG pipeline with control prompts and workflows. This approach actually gave us much better accuracy, stability, and transparency. And of course, right now, like, you know, whenever any tool comes to the market, it will be, like, raw.
100 00:14:20.650 ⇒ 00:14:27.319 Srinivas Saiteja Tenneti: So once we use it, use it, train it for models, and once we, like, keep on using it, we’ll get a hang of it.
101 00:14:27.550 ⇒ 00:14:40.399 Srinivas Saiteja Tenneti: Then we can, like, even the… there will be a few updates also for many of it, and then new models will be coming, like, let’s say there will be, like, few ones which… like, let’s say any… any tool is there, let’s say if you take,
102 00:14:40.690 ⇒ 00:14:53.649 Srinivas Saiteja Tenneti: I mean, if you take Pinecone only, let’s say Pinecone Vector Show. So, there will be new updates which will be coming, so because of that, it will be upgrading well. So, what I do is, like, I use in my personal projects.
103 00:14:53.650 ⇒ 00:15:03.740 Srinivas Saiteja Tenneti: So, once I get a hang of it, I will be, like, discussing that. If it really, like, you know, if it really changes the, like, if it will be easy, it will be making the workflow easier, then of course…
104 00:15:03.740 ⇒ 00:15:04.430 Samuel Roberts: Right, right.
105 00:15:04.430 ⇒ 00:15:06.409 Srinivas Saiteja Tenneti: that we’ll be using. Yeah.
106 00:15:06.710 ⇒ 00:15:15.100 Samuel Roberts: Okay, great, thank you. So we’re about halfway through, so I wanted to save some time to, answer any questions you have about the role, or Brain Forge, or anything like that.
107 00:15:15.280 ⇒ 00:15:19.079 Srinivas Saiteja Tenneti: Sure, sure. I want to, like, I have actually researched a lot about brain.
108 00:15:19.080 ⇒ 00:15:19.750 Samuel Roberts: Okay.
109 00:15:19.750 ⇒ 00:15:33.740 Srinivas Saiteja Tenneti: So, I have also, like, known about, automation techniques, I guess. You guys do, like, even AI… like, help the companies for AI automation and stuff. So, what I’m… just… it’s not a question, I’m just telling about, like, myself.
110 00:15:33.780 ⇒ 00:15:44.959 Srinivas Saiteja Tenneti: I’m good with Docker, and also, like, Kubernetes and MLflow, so these are, like, part and parcel of my, like, work which I do. And the question is that,
111 00:15:44.980 ⇒ 00:16:03.650 Srinivas Saiteja Tenneti: right now, like, I mean, Brainforce do many projects, like, you know, I have learned about Brainforce, like, it has been, like, you know, it helps companies, like, use AI, or, like, like, transfer to AI techniques, and also, like, help in automation, and also, like, help them to, like.
112 00:16:03.730 ⇒ 00:16:15.360 Srinivas Saiteja Tenneti: like, you know, what they say, increase their business productivity, or something. So, I want to know what are the really crazy, like, projects which you currently have? If you can share, of course.
113 00:16:15.360 ⇒ 00:16:28.680 Samuel Roberts: Yeah, yeah, let me think. So, so it’s interesting because we… we have kind of, you know, we have client work that we do, but we’re also doing a lot of internal work. So we’re trying to build, our own tooling, we have a…
114 00:16:28.860 ⇒ 00:16:46.829 Samuel Roberts: a little platform that we’ve built that, you know, ingests all the meetings we have, and so the transcripts are there, and so you can ask questions over that. That is all… we actually use Superbase for the back end of that and everything. So, you know, tools you’re mentioning, I’m like, yes, I know these tools, I’m glad to hear it.
115 00:16:47.500 ⇒ 00:17:05.080 Samuel Roberts: I’m trying to think what else we’ve done there. We’ve built a few tools using some voice things, we built a case study agent, so, you know, we, for sales and marketing, we have case studies that we generate, and that always takes time for the marketing person to interview someone, and so we had it, we built a tool to do the interviews.
116 00:17:05.079 ⇒ 00:17:09.580 Samuel Roberts: you know, I don’t know how… I mean, I can tell you kind of generally some of the client work, you know, we’ve done some…
117 00:17:09.579 ⇒ 00:17:21.260 Samuel Roberts: RAG chatbots as well. We’ve done some data using, like, MCP servers for getting data from various sources. We have done some basic
118 00:17:21.380 ⇒ 00:17:34.410 Samuel Roberts: I don’t say basic automation, but, you know, some people say, like, this is how I use Claude right now. I save all these prompts, and I copy and paste, and I copy and paste, and we say, okay, well, that we can definitely streamline for you.
119 00:17:34.410 ⇒ 00:17:36.489 Srinivas Saiteja Tenneti: You know, we really tried to…
120 00:17:36.580 ⇒ 00:17:55.759 Samuel Roberts: talk to, you know, various businesses, and learn what they’re trying to do, and how we think we can, you know, make things more productive, make it faster, make it cheaper, you know, rather than just kind of say, like, here’s what we offer, do you want it? It’s a much more of a,
121 00:17:56.300 ⇒ 00:18:02.389 Samuel Roberts: collaborative, you know, they think they need this, we think they need that, kind of, you know, where’s the solution here?
122 00:18:02.390 ⇒ 00:18:04.770 Srinivas Saiteja Tenneti: That is useful for them, and yeah, in that way.
123 00:18:04.770 ⇒ 00:18:12.570 Samuel Roberts: Yeah, exactly. So, yeah, I mean, it jumps around. Obviously, like, clients come and go, so the type of work kind of changes, but the internal work is stuff we’re constantly.
124 00:18:12.570 ⇒ 00:18:19.939 Srinivas Saiteja Tenneti: And it’s cool, because the internal stuff, we get to kind of experiment with those new things, like, it’s less, you know, not less critical, but it’s not…
125 00:18:19.940 ⇒ 00:18:25.070 Samuel Roberts: It’s client-facing, you’re not gonna lose business because it doesn’t work, we’re gonna make sure it, you know, we get to play with the new stuff, which is fun.
126 00:18:25.820 ⇒ 00:18:33.599 Srinivas Saiteja Tenneti: And, I think, like, for the voice and all, you will be using, like, TTH techniques, which is, like, text-to-speech, speech-to-text, and those kind of techniques, right?
127 00:18:33.630 ⇒ 00:18:38.839 Samuel Roberts: Yeah, we used… we did some of that, I forget exactly which model. One of the OpenAI ones had a real-time.
128 00:18:38.840 ⇒ 00:18:40.410 Srinivas Saiteja Tenneti: Whisper. OpenA Whisper, you mean?
129 00:18:40.500 ⇒ 00:18:43.270 Samuel Roberts: It wasn’t just Whisper, it was actually,
130 00:18:43.510 ⇒ 00:18:59.160 Samuel Roberts: one of the voice models that is an LLM, but had the voice input, so we didn’t have… because we thought we’d have to do a bunch of steps, and this one let us kind of just open a socket and talk right to it. And there was a little bit of weird stuff, because it would return without talk… it was, you know, lots of little…
131 00:18:59.160 ⇒ 00:18:59.579 Srinivas Saiteja Tenneti: in which case.
132 00:18:59.580 ⇒ 00:19:03.970 Samuel Roberts: But, yeah, it was good stuff. But yeah, we’ve been playing with Whisper, we’ve been playing with,
133 00:19:04.290 ⇒ 00:19:09.690 Samuel Roberts: I’m trying to think what else. We did some stuff with, image generation for a client.
134 00:19:09.690 ⇒ 00:19:10.040 Srinivas Saiteja Tenneti: That’s crazy.
135 00:19:10.040 ⇒ 00:19:11.049 Samuel Roberts: and a banana.
136 00:19:11.360 ⇒ 00:19:24.660 Srinivas Saiteja Tenneti: Yo, Nano Banana, yeah, that one, yeah, I tried one also. And, yeah, actually, like, you know, like, even for my personal project, I’m trying to do that, voice and video analysis. Okay.
137 00:19:24.660 ⇒ 00:19:25.380 Samuel Roberts: Yeah, yeah, right.
138 00:19:25.380 ⇒ 00:19:33.030 Srinivas Saiteja Tenneti: I’m trying to build that one using, like… even I’m doing for the dashboards using LangFuse, if you heard about it. Sure.
139 00:19:33.320 ⇒ 00:19:36.820 Srinivas Saiteja Tenneti: All are, like, a lot of tools I’m using, it is, like.
140 00:19:36.820 ⇒ 00:19:38.509 Samuel Roberts: Yeah, yeah, yeah, there’s so many, yeah.
141 00:19:38.510 ⇒ 00:19:42.300 Srinivas Saiteja Tenneti: Yeah, cursor ID, and also Claude as in, like, you know, to…
142 00:19:42.720 ⇒ 00:19:43.260 Samuel Roberts: Yep.
143 00:19:43.260 ⇒ 00:19:51.400 Srinivas Saiteja Tenneti: help me, Cloud code, so this is all I’m using, and it is also great, I would say, like, it is, like, those tools are pretty great.
144 00:19:51.450 ⇒ 00:20:10.160 Srinivas Saiteja Tenneti: And, yeah, one more question I have, if you don’t mind me asking. Sure, of course. Yeah, so this role involves building, like, AI products, right? Like, how clo… like, I want to ask, how do… how closely do, like, AI engineers, or not only AI engineers, like, usually engineers in the company work with customers or stakeholders during development?
145 00:20:10.620 ⇒ 00:20:18.709 Samuel Roberts: Yeah, so, we’re a relatively small AI team, so we have a data team and an AI team. The data team does more pipelines and all that sort of stuff.
146 00:20:19.630 ⇒ 00:20:22.479 Samuel Roberts: So the clients we work with, we work relatively closely with them.
147 00:20:22.860 ⇒ 00:20:24.860 Samuel Roberts: So we,
148 00:20:25.540 ⇒ 00:20:45.100 Samuel Roberts: you know, the engineers will talk to clients, we have a Slack channel that every client has their own external channel on, and so we like to be in communication with them, we don’t want to go two weeks without talking to them, and then all of a sudden we’ve built something in a different direction than they thought. So there’s a lot of back and forth, a lot of communication that way.
149 00:20:45.100 ⇒ 00:20:45.630 Srinivas Saiteja Tenneti: God.
150 00:20:45.800 ⇒ 00:20:46.170 Samuel Roberts: Sorry, go ahead.
151 00:20:46.170 ⇒ 00:20:49.100 Srinivas Saiteja Tenneti: No, no, I’m saying that’s actually, like, works well.
152 00:20:49.350 ⇒ 00:21:06.709 Samuel Roberts: Yeah, no, it’s great, because they, you know, we feel much more tied into them, they feel much more tied into us. We know what they’re looking for. When something’s not right, we can fix it quickly. You know, we check in a lot. At the minimum, it’s a weekly check-in, but usually it’s a every other day, everyday kind of thing.
153 00:21:07.000 ⇒ 00:21:07.620 Srinivas Saiteja Tenneti: Yeah.
154 00:21:07.740 ⇒ 00:21:08.260 Samuel Roberts: Yeah.
155 00:21:08.590 ⇒ 00:21:27.659 Srinivas Saiteja Tenneti: That actually, like, you know, when working with stakeholders or clients or, like, you know, customers, like, keep, like, having a scheduled meeting and work… getting their feedback is actually necessary. Otherwise, as you said, if they will ask for something, it will go in some other direction. Then ray tracing back, it will… it is like a messy thing.
156 00:21:27.700 ⇒ 00:21:39.439 Srinivas Saiteja Tenneti: So, yeah, and yeah, these were my very good questions, and I feel like these… what… these questions, what I wanted to know more about the company. And one more thing I wanted to know, like… Sure.
157 00:21:39.440 ⇒ 00:21:53.689 Srinivas Saiteja Tenneti: if, let’s say, fingers crossed, of course, if I join the team, then, let’s say, will I be, like, you know, having, like, something which I could build, and end of the day, I would be proud of it, like, this is my baby or something, which I could be, like,
158 00:21:53.920 ⇒ 00:21:54.440 Srinivas Saiteja Tenneti: We have…
159 00:21:54.440 ⇒ 00:21:58.580 Samuel Roberts: Yeah. Yeah, I think, I mean, we… you know, there’s a little bit of a…
160 00:21:58.710 ⇒ 00:22:04.230 Samuel Roberts: like, probationary time period, but during that time, you’re still working.
161 00:22:04.230 ⇒ 00:22:04.820 Srinivas Saiteja Tenneti: Perfect, sir.
162 00:22:04.820 ⇒ 00:22:19.150 Samuel Roberts: Yeah, so you’d be building things, there’s plenty of autonomy, you know, we don’t really, micromanage or babysit or anything like that, and so we have our morning check-ins, we kind of figure out the work and, you know, assign tickets and do that, but, like, it’s not…
163 00:22:19.300 ⇒ 00:22:33.029 Samuel Roberts: it’s not so much that it’s like, here’s how to do it, it’s oftentimes, this is what needs to get done, figure it out. We’ll talk about it, we’ll make a plan, you know, we can do whatever, you can experiment with things and come back, and yeah, it’s a lot of that sort of stuff. So, you know, I think…
164 00:22:33.200 ⇒ 00:22:35.509 Samuel Roberts: Depending on… Oh yeah, go ahead.
165 00:22:35.660 ⇒ 00:22:45.240 Srinivas Saiteja Tenneti: Sorry, sorry, sorry. So, usually, like, what, you know, like, I really love… I’m actually excited to be a part, to be a part of this company, because…
166 00:22:45.300 ⇒ 00:22:56.169 Srinivas Saiteja Tenneti: what are all the work, like, you know the company is doing, and what all the… what all you have, like, explained to me. So, everything, like, you know, aligns with me, and apart from that.
167 00:22:56.170 ⇒ 00:23:06.510 Srinivas Saiteja Tenneti: at one stage, you know, I want to be, like, an, you know… there is a dream of me, if you don’t mind, so I want to be, like, you know, a really critical member of the team.
168 00:23:06.510 ⇒ 00:23:18.570 Srinivas Saiteja Tenneti: Like, he really… he owns the product, or he owns the pipeline, or something, but I’m really, like, you know, looking forward to it. Because that one, so, you know, gives me goosebumps, or, you know, makes me crazy.
169 00:23:18.650 ⇒ 00:23:35.409 Srinivas Saiteja Tenneti: I’m, like, a geek. I’m not, like, a geek geek, but I’m a geek of, you know, I usually do things, but I also, like, in the spare time, I check the newsletters, try different new products. If it is good, I will, like, have a discussion with the team and discuss it and, you know, implement.
170 00:23:35.410 ⇒ 00:23:42.580 Srinivas Saiteja Tenneti: I’m, like, constantly, like, you know, I… you know, I try to, like, grab whatever it is there, understand it, learn.
171 00:23:42.580 ⇒ 00:23:46.750 Srinivas Saiteja Tenneti: Do certifications, and also learn about a lot of new topics.
172 00:23:46.760 ⇒ 00:23:50.830 Srinivas Saiteja Tenneti: So I keep on doing that, and it was great speaking to you, actually, like.
173 00:23:50.830 ⇒ 00:24:08.780 Samuel Roberts: Yeah, you as well, I really appreciate the time. So yeah, I think next steps, you know, if I bring this back to the team and everyone’s good to go, you’d do a second interview that’s a little more role-focused and technical, and then if that goes well, there’s a third panel interview and, a tech challenge kind of thing that we would be able to talk about.
174 00:24:08.780 ⇒ 00:24:22.089 Samuel Roberts: But we like to move relatively quickly. You know, I think I’ve been saying the biggest hurdle is just scheduling this kind of time and making sure we have it, because, you know, if things go well, you could be scheduling the next one by the end of the week kind of thing, so we don’t like to drag this out. Of course, of course.
175 00:24:22.520 ⇒ 00:24:24.659 Samuel Roberts: You should be hearing back one way or another, yeah.
176 00:24:24.660 ⇒ 00:24:31.849 Srinivas Saiteja Tenneti: Sure, sure. And I’m, like, fingers crossed. I’m… I’m really crazy about this company, Brainforge, because…
177 00:24:31.880 ⇒ 00:24:45.619 Srinivas Saiteja Tenneti: You know, I have… it’s not about the work, and actually, I have researched it a lot. Okay. Like, I have gone through the brain forge, I have gone through the linkering of… if you don’t mind, I have gone through lingering of all the people who work there.
178 00:24:45.620 ⇒ 00:24:46.489 Samuel Roberts: Oh, wow, yeah.
179 00:24:46.490 ⇒ 00:24:49.799 Srinivas Saiteja Tenneti: It’s not about stalking, but I really like research a lot.
180 00:24:50.400 ⇒ 00:25:01.290 Srinivas Saiteja Tenneti: It’s crazy, the workspace is crazy, and of course, the job opportunity is remote. It is like, you know… I felt like, you know, you give the proper, balance to the people who are working.
181 00:25:01.850 ⇒ 00:25:09.819 Srinivas Saiteja Tenneti: you let them, like, have the proper space, and then it will really help, like, you know, we deliver the best output, so I really, like…
182 00:25:09.940 ⇒ 00:25:14.550 Srinivas Saiteja Tenneti: I really appreciate that, and thank you for the opportunity. I’m really grateful for that.
183 00:25:15.020 ⇒ 00:25:22.069 Samuel Roberts: Great, yeah. So, yeah, you should hear back one way or the other soon, and then, yeah, hopefully I’ll look forward to seeing you in the future.
184 00:25:22.070 ⇒ 00:25:32.310 Srinivas Saiteja Tenneti: Sure, sure. Thank you, and I hope I get next steps, and of course, I’m praying for it, and… and thank you for the opportunity, and of course, I’m grateful to you, and have a great day.
185 00:25:32.310 ⇒ 00:25:33.989 Samuel Roberts: You as well.
186 00:25:34.530 ⇒ 00:25:34.940 Srinivas Saiteja Tenneti: Thank you.
187 00:25:34.940 ⇒ 00:25:35.580 Samuel Roberts: Easy.
188 00:25:35.840 ⇒ 00:25:36.610 Srinivas Saiteja Tenneti: Thank you.