Meeting Title: Brainforge Interview w- Sam Date: 2026-04-03 Meeting participants: Vimarsh Patel, Samuel Roberts
WEBVTT
1 00:03:37.340 ⇒ 00:03:38.130 Vimarsh Patel: Good morning.
2 00:03:39.000 ⇒ 00:03:42.399 Samuel Roberts: Good morning. Second, sorry, just getting everything going here.
3 00:03:44.080 ⇒ 00:03:45.639 Samuel Roberts: Okay, there we go, okay.
4 00:03:46.220 ⇒ 00:03:47.140 Samuel Roberts: How are you?
5 00:03:47.730 ⇒ 00:03:49.340 Vimarsh Patel: I’m doing good, how about you?
6 00:03:49.650 ⇒ 00:03:52.980 Samuel Roberts: Doing well, doing well. Thanks for taking the time today.
7 00:03:53.300 ⇒ 00:03:54.339 Vimarsh Patel: Let’s see…
8 00:03:54.680 ⇒ 00:03:55.210 Samuel Roberts: Keep doing that.
9 00:03:55.210 ⇒ 00:03:56.080 Vimarsh Patel: the trends.
10 00:03:56.080 ⇒ 00:04:09.900 Samuel Roberts: Of course, of course. My name is Sam Roberts, I’m the AI tech lead here at Brainforge. So, I have some questions here. I want to make sure we leave time for you to ask any questions, and then,
11 00:04:10.170 ⇒ 00:04:20.030 Samuel Roberts: But to start, I’d love for you to just… I’ve seen your loom and your LinkedIn, but I’d love to get just a quick elevator pitch kind of intro of yourself and your background.
12 00:04:20.350 ⇒ 00:04:40.179 Vimarsh Patel: Sure, sure. I am Vimarsh. My background is AI engineer, that’s what this role attracted me so much. But I would say, like, before that, I was kind of a bit of a full-stack engineer as well, but I am much more stronger on the back-end side, building APIs and connecting it with the front end. Apart from that,
13 00:04:40.420 ⇒ 00:04:44.180 Vimarsh Patel: in, like, I do love, like.
14 00:04:44.680 ⇒ 00:04:56.779 Vimarsh Patel: AI that’s been coming up, like, agents and track systems are the new norm to be incorporated into the software, so that’s what I’ve been trying to grow and build my skillset as. And…
15 00:04:56.860 ⇒ 00:05:07.579 Vimarsh Patel: I am, like, looking to grow in my career next, and that is why this role attracted me, and then I just went ahead and applied for it, so that is my…
16 00:05:07.580 ⇒ 00:05:08.200 Samuel Roberts: Great.
17 00:05:08.410 ⇒ 00:05:11.879 Vimarsh Patel: There’s extra background, but yeah, happy to answer more questions as you ask me.
18 00:05:11.880 ⇒ 00:05:27.020 Samuel Roberts: Yeah, I’m sure we’ll get into it a little bit. So, let’s, let’s talk about, I mean, I heard a little bit in the Loom, or the video, but let’s talk about an LLM-based feature that you shipped to production, and the problem that it solves.
19 00:05:27.320 ⇒ 00:05:27.950 Samuel Roberts: Specifically.
20 00:05:27.950 ⇒ 00:05:45.750 Vimarsh Patel: Yes, so currently, like, I am building, like, a reconciliation system in my current role, where, we are… I am trying to, like, reduce the manual task the HR has. Basically, we have a three-way… like, three data sources. One is ADP, QuickBooks, and Outlook.
21 00:05:45.800 ⇒ 00:06:00.930 Vimarsh Patel: So, we are trying to match the timesheets across the three, and then generate invoices for the client. So, based on that, the HR has to manually verify the timesheets from EDP and the timesheets sent by the client.
22 00:06:00.980 ⇒ 00:06:08.099 Vimarsh Patel: And they have to, like, verify, and then they charge the client based on that. It is quite a manual task, and…
23 00:06:08.100 ⇒ 00:06:08.430 Samuel Roberts: Yes.
24 00:06:08.430 ⇒ 00:06:24.959 Vimarsh Patel: For the number of employees that the company has, the HR team has, like, it’s a very gruesome and tiresome task for them to open every file, go through that, and then generate the invoice. Sometimes, if there’s a discrepancy, then they have to go back and forth between the client and the employee.
25 00:06:24.960 ⇒ 00:06:32.200 Vimarsh Patel: So, this is one task that I’m helping them challenge. Obviously, we are doing… I am implementing it in layers.
26 00:06:32.200 ⇒ 00:06:47.550 Vimarsh Patel: Because you cannot go from 0 to 1 in one go, you have to take it step by step, and in production especially, there are a lot of edge cases that keep coming up. You have to keep that in mind as well. So, currently, what I’m trying to do is I have connected all the… like, I have connected Outlook and the ADP,
27 00:06:47.550 ⇒ 00:07:02.399 Vimarsh Patel: And I am extracting the timesheets, one entered by the employee and one sent by the client, and I’m extracting the data using a Python library, and I’m using the LLM for structuring that data in a JSON, and then inputting it into an SQL for reconciliation, basically.
28 00:07:02.400 ⇒ 00:07:16.489 Vimarsh Patel: So that is my step one, and that is where I am using the LLM. So at least one part has been reduced, like, they don’t have to manually go and open all the file and check. They can just search the employee, they can get the week, like, maybe week ending for, like, last week, and they can just see the approved
29 00:07:16.530 ⇒ 00:07:23.469 Vimarsh Patel: hours from the client, and the client… and the, hours entered by the… so from the ADP. Okay.
30 00:07:23.470 ⇒ 00:07:28.890 Samuel Roberts: So, just one sec… so, so if you’re using a library to get the data.
31 00:07:29.390 ⇒ 00:07:29.980 Vimarsh Patel: Yes.
32 00:07:29.980 ⇒ 00:07:34.489 Samuel Roberts: The LLM is helping you structure the data before you put it into the database?
33 00:07:34.490 ⇒ 00:07:42.790 Vimarsh Patel: Yes, because, like, if I extract it from PyPDF or, like, PIMU PDF, like, it’s, like, in a very, in a bad way, like, it’s very…
34 00:07:42.790 ⇒ 00:07:43.180 Samuel Roberts: Yep.
35 00:07:43.410 ⇒ 00:07:44.059 Vimarsh Patel: I just wasn’t.
36 00:07:44.060 ⇒ 00:07:45.639 Samuel Roberts: I’m sure where the quality was there, yeah, so I understand.
37 00:07:45.640 ⇒ 00:07:46.179 Vimarsh Patel: Yeah, exactly.
38 00:07:46.180 ⇒ 00:07:46.780 Samuel Roberts: Oh, I’m there.
39 00:07:47.090 ⇒ 00:07:59.209 Vimarsh Patel: And, like, the reconciliation task is very deterministic. Like, if A equals to B, then, yes, it is approved. I don’t need an L&M for that. Like, why to increase the… why to increase the cost and, like, base the tokens for a.
40 00:07:59.210 ⇒ 00:07:59.680 Samuel Roberts: Right.
41 00:07:59.680 ⇒ 00:08:12.709 Vimarsh Patel: task. So, like, right now, the LLM part is only for this, but I have other plans as we go forward. Maybe I can introduce an agent. Like, if there’s a discrepancy in the reconciliation, then maybe the agent can
42 00:08:12.970 ⇒ 00:08:18.390 Vimarsh Patel: take care of it. It can put in a… it can call the human reviewer, and then maybe, like.
43 00:08:18.960 ⇒ 00:08:33.630 Vimarsh Patel: to put in a notification or an alert, and then for the HR to review that discrepancy, so maybe the agent can help in such a way, so… Okay. So that’s what my Layer 2 will be, like, right.
44 00:08:33.789 ⇒ 00:08:42.379 Vimarsh Patel: And then my Layer 3 would be connecting it with the QuickBooks. So, once it is approved, it can automatically generate the invoice, and the agent can send it to the
45 00:08:42.750 ⇒ 00:08:56.180 Vimarsh Patel: to the client, once the human reviewer part is done, obviously, because we cannot let it completely automate it, like, we need at least one human reviewer to make sure that they are charging correctly, like, the rate is correct, and everything here and there. So…
46 00:08:56.300 ⇒ 00:08:58.559 Vimarsh Patel: Okay. This is what I’m currently trying to implement.
47 00:08:58.670 ⇒ 00:08:59.380 Vimarsh Patel: Perfect.
48 00:08:59.380 ⇒ 00:09:06.540 Samuel Roberts: Thank you, thank you. So I would say, my next question is kind of… you talked about what you’ve built so far and what you’re looking to build.
49 00:09:06.540 ⇒ 00:09:06.900 Vimarsh Patel: Yes.
50 00:09:06.900 ⇒ 00:09:10.129 Samuel Roberts: Within the AI stack, have you spent the most time
51 00:09:10.550 ⇒ 00:09:16.869 Samuel Roberts: In general, like, where do you feel… like, building, like, production stuff versus experimenting and playing around? Like, where do you.
52 00:09:16.870 ⇒ 00:09:17.580 Vimarsh Patel: So…
53 00:09:17.580 ⇒ 00:09:18.580 Samuel Roberts: comfortable, yeah.
54 00:09:18.780 ⇒ 00:09:41.059 Vimarsh Patel: where do I feel most comfortable is… one is obviously connecting the LLM with the software, like the LLM API with the software. Like, as I said, I’m strong with the backend as well, so connect… building APIs and connecting with the front end, and the other parts as well, like, connecting the data sources with the backend, moving the data here and there, what, like, what I’m sending to the LLM, what I’m getting back, and
55 00:09:41.210 ⇒ 00:09:57.339 Vimarsh Patel: all that parts are where I thrive, and that is where I’m most comfortable as well. Apart from that, recently I’ve also tried to build RAG pipelines and trunking strategies, so that’s where also I’m trying to grow as an AI engineer, because.
56 00:09:57.340 ⇒ 00:09:57.810 Samuel Roberts: Sure.
57 00:09:57.810 ⇒ 00:10:04.379 Vimarsh Patel: But that is also the next part, and I have done some side projects here and there to build agents as well, so…
58 00:10:04.380 ⇒ 00:10:04.850 Samuel Roberts: Great.
59 00:10:04.850 ⇒ 00:10:11.059 Vimarsh Patel: But the agent part, I am still, like, getting there, but I’m much more comfortable with RAG and the other data pipelines.
60 00:10:11.060 ⇒ 00:10:16.099 Samuel Roberts: Okay, great. And what tools have you been using for RAG? Like, what do you go to normally?
61 00:10:16.470 ⇒ 00:10:30.329 Vimarsh Patel: So, my go-to vector DB would be ChromaDB, because, like, for testing out, and that is really good, or even PGVector. PGVector is good. For chunking, I usually,
62 00:10:30.440 ⇒ 00:10:46.460 Vimarsh Patel: have different strategies, depending on how big the document is and, like, how much the limit is allowed. In terms of backend, I try to use Python and FastAPI, or even Node.js is also what I’m comfortable with. Okay.
63 00:10:46.460 ⇒ 00:11:06.129 Vimarsh Patel: I have also used some automation platforms, like N810. If it’s a simple workflow, then I can just build it in N810 and then expose it using a webhook, so the code… so my code can call that N810 automation instead of, you know… like, if it’s a simple, basic automation, then I don’t need to code all the things. It’s much easier if you…
64 00:11:06.380 ⇒ 00:11:07.610 Vimarsh Patel: Yeah, so…
65 00:11:07.610 ⇒ 00:11:15.289 Samuel Roberts: Something we do here is sometimes N8N for a prototype, and then move to code for hardening. Great. Okay, so… Oh, sorry, go ahead.
66 00:11:15.290 ⇒ 00:11:16.600 Vimarsh Patel: No, no, no, no, no.
67 00:11:16.600 ⇒ 00:11:28.319 Samuel Roberts: No, you’re good, okay. The next thing I want to kind of talk about is maybe a little less… not less technical, but, we deal with clients, we deal with other people in the company that hear a lot about AI, but might not be
68 00:11:28.320 ⇒ 00:11:39.750 Samuel Roberts: technical people, what… how do you go about explaining the limitations of LLMs and the agents and all this other technology to people that, one.
69 00:11:39.830 ⇒ 00:11:50.189 Samuel Roberts: don’t understand the depths of it, but two, hear all about it in the news and everything, so they might not know exactly, but they think they know. So, how would you go about it? How do you think about that?
70 00:11:50.730 ⇒ 00:11:59.810 Vimarsh Patel: Okay, so I think I can take the same example with the 3-way reconciliation system that I’m building, so that would be much easier. So, like, the… like, the…
71 00:11:59.890 ⇒ 00:12:17.730 Vimarsh Patel: like, the company MD or, like, the CEO, like, I was just talking to him, and he was like, I want to build, like, an automation for this, like, for reconciliation, and, like, he’s like, I want it in a way, like, like an AI agent work. But, like, when I understood what he… like, the pain point and the workflows are that he wants, like, I got to learn that, like, like, an AI agent is, like, an…
72 00:12:17.790 ⇒ 00:12:24.350 Vimarsh Patel: It’s a big overkill for the initial task. I mean, like, if you just want to do A equals to B, then you don’t need an AI agent for that. It’s like a…
73 00:12:24.350 ⇒ 00:12:24.750 Samuel Roberts: Right, right.
74 00:12:24.990 ⇒ 00:12:34.869 Vimarsh Patel: It’s just, like, it’s just going to waste his money, time, energy, everything for people who’s going to build it, and these people’s gonna, you know, someone who’s going to use it as well, because it’s going to be so much more difficult to debug.
75 00:12:34.870 ⇒ 00:12:47.650 Vimarsh Patel: So I had to, like, I didn’t tell him no exactly, but I told him that, yes, automation can be done, but you don’t need exactly an AI agent for that. I made him realize that where an AI agent would be useful in this scenario.
76 00:12:47.650 ⇒ 00:12:57.390 Vimarsh Patel: once we build it, and how to automate the process currently without AI as well, because people think that automation means AI, but that is not the exact case every day.
77 00:12:57.390 ⇒ 00:12:57.760 Samuel Roberts: Right.
78 00:12:57.760 ⇒ 00:13:05.959 Vimarsh Patel: I think, sometimes, non-technical people are mis… misinterpret, or, like, they are, like, they…
79 00:13:06.070 ⇒ 00:13:11.260 Vimarsh Patel: understand something, and, like, they believe that that is your life. So I think our job here would be to
80 00:13:11.410 ⇒ 00:13:22.170 Vimarsh Patel: to tell them, yes, it can be done, what they want is doable, but not in the exact way that they want. Or maybe what they understood, and, like, what they want can be two different things sometimes.
81 00:13:22.170 ⇒ 00:13:22.530 Samuel Roberts: Yeah.
82 00:13:22.530 ⇒ 00:13:23.010 Vimarsh Patel: So…
83 00:13:23.010 ⇒ 00:13:23.669 Samuel Roberts: Totally, totally.
84 00:13:23.670 ⇒ 00:13:26.520 Vimarsh Patel: I think maybe we, I would, like,
85 00:13:27.320 ⇒ 00:13:45.630 Vimarsh Patel: take an approach where I make them understand what they are, like, what they… what technical term they are using, how they are, like, if it… if it blends well, then yes, obviously, like, if you want an AI agent for this work, and if it’s suitable, then yes, we can definitely go and make them understand how it works. But, like, if it’s an overkill, then it is our job as well
86 00:13:45.630 ⇒ 00:13:57.339 Vimarsh Patel: to make them realize that it is an… like, you are making use of your resources in a… in the wrong way. Like, so… because, obviously, they are our client, they are going to be paying us, we don’t want to waste their money as well. Exactly.
87 00:13:57.340 ⇒ 00:13:59.389 Samuel Roberts: Exactly. So, yeah, I think… Great. Okay.
88 00:13:59.890 ⇒ 00:14:01.480 Vimarsh Patel: This, this, this would be my approach.
89 00:14:01.480 ⇒ 00:14:06.820 Samuel Roberts: Perfect, no, that sounds… that sounds great. Let’s talk about, just…
90 00:14:07.230 ⇒ 00:14:26.679 Samuel Roberts: what is some… like, when something goes wrong with an AI system, or, you know, the LLM giving bad results, and, or someone misunderstands what it’s capable of and thinks they’re bad results, like, where do you take responsibility? Where do you figure out, or try to explain the limit? Like, where… how do you balance that?
91 00:14:27.340 ⇒ 00:14:38.659 Vimarsh Patel: So, like, I have one follow-up before I answer this. Like, are you saying that if we get stuck while building the code, or, like, if someone else is using our product and they are getting stuck with the… that, like.
92 00:14:38.660 ⇒ 00:14:57.739 Samuel Roberts: I would say if someone else is using a product you built, and they either misunderstand how it works, so they think it’s giving bad outputs, or they’re asking, you know, they’re using it in a way that’s not intended, or it underperforms, actually, so there’s, you know, there’s kind of two A’s here, but, like, where do you…
93 00:14:58.010 ⇒ 00:15:08.470 Samuel Roberts: How do you go about, like, taking responsibility, or explaining the system, or where’s that line for you, where the user… all the user cares about is getting the right result, and if they’re not getting it…
94 00:15:08.850 ⇒ 00:15:10.179 Samuel Roberts: It can be a few different reasons.
95 00:15:10.180 ⇒ 00:15:10.750 Vimarsh Patel: Obviously.
96 00:15:10.750 ⇒ 00:15:11.400 Samuel Roberts: How about that?
97 00:15:11.810 ⇒ 00:15:30.630 Vimarsh Patel: Yeah, I mean, so obviously, like, we cannot demean the user, like, the user is always right, like, the user, the customer, the client, they’re always right, and, like, it is our job to get what they want. In the first scenario, like, if they’re interpreting it, or, like, if they’re using it in the wrong way, then maybe, I think, at the, like, whatever platform or, like, whatever product that we’re giving out, maybe we can
98 00:15:30.630 ⇒ 00:15:35.489 Vimarsh Patel: give a small walkthrough or the demo. Initially, like, a lot of tools do that when you
99 00:15:35.490 ⇒ 00:15:39.090 Vimarsh Patel: Like, log in for the first time, they give you a walkthrough of the whole
100 00:15:39.090 ⇒ 00:15:52.089 Vimarsh Patel: all the features. So maybe we can do a small something like that in the beginning. Like, if it’s a first-time user, we can put that. Or maybe we can, put in a small video on our website or on the app store, where, like, or just, like,
101 00:15:52.240 ⇒ 00:15:55.409 Vimarsh Patel: Like, how to use the main app, and what it is
102 00:15:56.150 ⇒ 00:16:15.380 Vimarsh Patel: That would be the first approach, because, like, it’s easy for the user to use if they know what capabilities the app has. Secondly, like, if it is actually underperforming, and if it is giving out the wrong outputs, then I think feedback is obviously very important. You are not going to be able to
103 00:16:15.380 ⇒ 00:16:23.990 Vimarsh Patel: make the product 100% right in the first go. I think user feedback is definitely the most important aspect here, because the output is actually wrong, so…
104 00:16:24.040 ⇒ 00:16:37.430 Vimarsh Patel: We take the feedback from the user, we… we log it, we understand what feature went wrong, or, like, what part of the output was wrong. We try to, like, reverse engineer it, go back, understand, like,
105 00:16:37.430 ⇒ 00:16:50.179 Vimarsh Patel: like, what did the… like, obviously, we log everything, like, when you’re making a production tool, we are going to log every single thing that the user did, like, if they logged in, if they clicked this, if they clicked that, they’re going to know everything, so we can just go back
106 00:16:50.180 ⇒ 00:16:58.830 Vimarsh Patel: From the output to what the user had clicked on, and then we see where the output got mismatched, and maybe then, based on the feedback, we can try to, you know, improve our product.
107 00:16:59.400 ⇒ 00:17:07.139 Samuel Roberts: Great, thank you. Both, both good answers. All right, let’s, let’s step back a little bit, maybe, and say,
108 00:17:07.510 ⇒ 00:17:16.319 Samuel Roberts: if you had 6 months of no obligations, what would you do? What would you spend 6 months if you just could…
109 00:17:16.460 ⇒ 00:17:18.559 Samuel Roberts: Do whatever you wanted for 6 months. What would you spend the time.
110 00:17:18.569 ⇒ 00:17:20.989 Vimarsh Patel: Six months in my personal life, or in my professional life.
111 00:17:21.329 ⇒ 00:17:36.669 Samuel Roberts: However, I mean, however you really want to answer, I’m just curious, like, what I, you know, I’m trying to get to, I guess, you know, if you had no obligations, would you… what would you want to work on? What would you want to do with your time? You know, feel… I mean, feel free to answer however you want. I’m just looking to understand a little more about you, yeah.
112 00:17:36.670 ⇒ 00:17:38.419 Vimarsh Patel: Sure, absolutely. So, like,
113 00:17:38.490 ⇒ 00:17:55.219 Vimarsh Patel: how I look at it is, like, like, everyone is going towards AI, so that’s completely fine. So, like, what I would like to do is, like, obviously, it is very easy for the big enterprises to implement AI, and they can, like, they have enough capital to, like, throw it all away if it doesn’t work.
114 00:17:55.220 ⇒ 00:18:16.190 Vimarsh Patel: But for small-medium businesses, or, like, small-medium enterprises, like, for people, for someone who’s, like, from 50 to 100 people, employees, it’s very difficult for them to, like, implement AI also, and… or if someone… or if companies not from a technical background, like, if it’s a manufacturing or a logistics company, they are… it’s much difficult for them, because they don’t have a specialized, team.
115 00:18:16.190 ⇒ 00:18:19.199 Vimarsh Patel: for software. They outsource everything, so…
116 00:18:19.200 ⇒ 00:18:28.119 Vimarsh Patel: What I would like to do is, like, maybe go to a place, maybe a Tier 2 city or somewhere, like, outside of the United States, maybe back in India, my home country.
117 00:18:28.250 ⇒ 00:18:39.849 Vimarsh Patel: like, not even the metro cities, like, not Delhi or Mumbai, but, like, Tier 2, maybe something like Nagpur, or maybe something like that, where, these companies are actually, like.
118 00:18:39.850 ⇒ 00:18:54.009 Vimarsh Patel: are still on the very paperwork kind of a system, they are not even that digitalized. They might have a website, they might have an app, but still, like, people are still not using it, so maybe try and understand their work process, and at the same time, I get to explore a new city as well. I like to travel.
119 00:18:54.010 ⇒ 00:19:03.309 Vimarsh Patel: So, maybe I can do both. So, maybe I try to understand some of these industries where AI is not that implemented, understand how unstructured that data is. So, from my point of view.
120 00:19:03.310 ⇒ 00:19:13.320 Vimarsh Patel: how I see is that even if we are using AI, I think still 80% of the data that exists is very unstructured for the AI, like… like, your AI model is only as good as the data you provide, so…
121 00:19:13.320 ⇒ 00:19:13.680 Samuel Roberts: Right.
122 00:19:13.680 ⇒ 00:19:30.100 Vimarsh Patel: like, a lot of these car companies, industries have 80% of their data unstructured, so I think I go there, understand their workflow, how unstructured their data is, how bad is it, and then come up with a way to structure that, and then maybe help that company implement AI and, like, increase their efficiency. Maybe, like, so…
123 00:19:30.220 ⇒ 00:19:40.110 Vimarsh Patel: this is not a spot… idea on the spot, like, I have been thinking about this for quite some time, about, like, how unstructured, too structured is causing a big
124 00:19:40.250 ⇒ 00:19:59.609 Vimarsh Patel: I can see it’s a big gap, I can definitely feel it as well, because I was… I was working on this three-way project, and I can see, like, the invoices are here, the approval sheets are here, and, like, it’s so, like, and it is a consulting company, but, like, if you go to a very hardcore business, like a plumbing business, or maybe, like, a laundry business.
125 00:19:59.840 ⇒ 00:20:11.960 Vimarsh Patel: they are going to be here and there, like, all the invoices and everything, they’re going to have the small receipts being printed out, or like in even a retail store. So, all these industries have so much unstructured data.
126 00:20:12.250 ⇒ 00:20:12.630 Samuel Roberts: Totally.
127 00:20:12.630 ⇒ 00:20:13.460 Vimarsh Patel: Oh, yeah? Great.
128 00:20:13.460 ⇒ 00:20:14.010 Samuel Roberts: Yeah, I love that.
129 00:20:14.010 ⇒ 00:20:14.630 Vimarsh Patel: Every book.
130 00:20:15.140 ⇒ 00:20:26.189 Samuel Roberts: That sounds really good. I have a few more questions, but we’re about halfway, so I want to just make sure we leave time for you to ask anything about the role, or Brainforge, or anything you have for me.
131 00:20:26.190 ⇒ 00:20:27.030 Vimarsh Patel: Yes.
132 00:20:27.030 ⇒ 00:20:28.669 Samuel Roberts: I’m happy to answer those now. Sure, sure, sure.
133 00:20:29.020 ⇒ 00:20:35.240 Vimarsh Patel: Yep, yep. So, like, obviously I know about Brain Phase, my recruiter was, like, they gave me a good…
134 00:20:35.360 ⇒ 00:20:51.630 Vimarsh Patel: lot of information. What I would like to ask is about the… about the team, probably, and, like, the multiple… and, like, I was also mentioned that, like, a team might serve multiple clients, so, like, how does that work as well, and, like, how… how are we assigned each role, and how it goes about that?
135 00:20:51.920 ⇒ 00:20:58.189 Samuel Roberts: Yeah, yeah, so the way we’re kind of organized, so historically, Brainforge started as more of a data consultancy.
136 00:20:58.380 ⇒ 00:21:12.479 Samuel Roberts: And so there’s a data engineering team, and then from that, building internal tooling, there were a couple AI engineers that were doing that, and then clients started asking about that, as you can imagine, so an AI team kind of was spun
137 00:21:12.480 ⇒ 00:21:24.089 Samuel Roberts: spun out of that that does internal, external stuff. And so, we’re a relatively small team. I mean, the company’s not huge yet, we’re growing now, but we have a few, there’s…
138 00:21:24.540 ⇒ 00:21:36.869 Samuel Roberts: 3 or 4 of us that are working on AI, projects for clients, and so, the way it works is, you know, the sales team goes out, and right now, we’re kind of in the process of trying to,
139 00:21:36.870 ⇒ 00:21:47.620 Samuel Roberts: kind of standardize our offerings. So, in the data side, you can imagine there’s, like, you know, setting up pipelines and doing modeling, and the AI side, we’re kind of like a full-stack development shop right now.
140 00:21:47.630 ⇒ 00:22:06.530 Samuel Roberts: Where we do that with AI, and so we’re trying to figure out, like, what are our best offerings? But that being said, when something comes up, when a client has an AI idea, and we can… we kind of work with the salespeople to figure that out, we… we have a few… so you can be on a couple different clients, potentially, plus some internal work. It kind of depends how…
141 00:22:06.530 ⇒ 00:22:12.230 Samuel Roberts: you know, much work there is per client on any given time, where people bounce around. I don’t love
142 00:22:12.230 ⇒ 00:22:28.580 Samuel Roberts: people going in and out of a client so much, but, like, once you’re on a client, you have all that knowledge and everything, you want to keep that going. So usually there’s, like, one or two clients, and internal is kind of a client we think about a little bit, too. Yeah. But, yeah, we’re… we’re kind of a team…
143 00:22:28.580 ⇒ 00:22:32.790 Samuel Roberts: That’s spread across different… different workflows, so it’s… it’s interesting.
144 00:22:32.790 ⇒ 00:22:34.770 Vimarsh Patel: It’s like a startup inside a company.
145 00:22:34.980 ⇒ 00:22:38.479 Samuel Roberts: A little bit, a little bit, but it’s also because we have the client work, we’re working on.
146 00:22:38.480 ⇒ 00:22:38.820 Vimarsh Patel: Yeah!
147 00:22:38.820 ⇒ 00:22:51.540 Samuel Roberts: So it’s a startup, but we’re also, like, you know, we’re not all building towards the same one thing all the time, so it’s… we have to, like, context shift and figure that out, but the team is, like I said, it’s about four of us right now.
148 00:22:51.560 ⇒ 00:22:58.999 Samuel Roberts: And so we’re, we’re pretty, you know, active, we’re pretty busy. There’s a lot of work everywhere, as you can imagine.
149 00:22:59.000 ⇒ 00:22:59.640 Vimarsh Patel: That’s great.
150 00:22:59.870 ⇒ 00:23:04.990 Samuel Roberts: But yeah, so we bounce around, and, like, clients come and go as different projects rise and, you know, start up, and…
151 00:23:04.990 ⇒ 00:23:05.960 Vimarsh Patel: Yeah, yeah.
152 00:23:05.960 ⇒ 00:23:20.250 Samuel Roberts: finish, so it varies quite a bit, so far. We’ve built things from just simple, N8N flows to very complex N8N flows that we have migrated into some TypeScript code using Mastra, if you’ve heard of that package.
153 00:23:20.250 ⇒ 00:23:21.149 Vimarsh Patel: No, not Maestro.
154 00:23:21.150 ⇒ 00:23:23.690 Samuel Roberts: It’s like a Lang chain, but in TypeScript.
155 00:23:23.690 ⇒ 00:23:24.040 Vimarsh Patel: Okay.
156 00:23:24.040 ⇒ 00:23:35.320 Samuel Roberts: And so, we do a little bit of Python. I have a big TypeScript background, so I feel more comfortable with that. And plus, when we’re building UIs and websites that people are interacting with, it’s good. I like the full stack.
157 00:23:35.320 ⇒ 00:23:45.680 Samuel Roberts: TypeScript, so when we need it, we go to Python, but right now, a lot of TypeScript. But the other thing I will talk about with you a little bit more is coding agents and how we use them. So, you know, things have changed so much.
158 00:23:45.680 ⇒ 00:23:46.050 Vimarsh Patel: Yeah, yeah.
159 00:23:46.050 ⇒ 00:23:49.419 Samuel Roberts: to that maybe a little bit, but, yeah, I mean, we,
160 00:23:49.690 ⇒ 00:24:13.480 Samuel Roberts: we’ve done some stuff like that, we’ve done some basic automations where it’s just someone was doing the same thing in Claude over and over again, and we said, well, we can automate that system for you, it’s, you know, it’s… the AI side is just what we’re automating, really, but other ones where it’s a RAG system, a chat application, other ones where it’s MCP, getting data, analyzing data, that sort of stuff. So really, whatever clients’ needs are, we try to figure out
161 00:24:13.480 ⇒ 00:24:17.600 Samuel Roberts: Very much what you talked about. I think I come from a very startup.
162 00:24:17.620 ⇒ 00:24:28.830 Samuel Roberts: more product-focused background. So, when I think of stuff, I’m not necessarily just like, alright, give me the specs, and we’ll go build it. It’s working with the client to understand what they need, and working with the…
163 00:24:28.830 ⇒ 00:24:40.079 Samuel Roberts: kind of, the CSOs, we call them the client success, and so they… they’re the ones that communicate with the client, and we communicate with them, and we all kind of work together to figure out what needs to happen, and then we can build. That’s kind of how the team works.
164 00:24:40.750 ⇒ 00:24:43.750 Vimarsh Patel: And, like, roughly, how long do the engagement with the client go on?
165 00:24:44.050 ⇒ 00:25:03.759 Samuel Roberts: Oh, boy. It can be anything from… I mean, one client we’ve been working on, for probably about a year in different forms, you know, as we’ve built the RAG system, we’ve extended it, we’ve moved it around, we’ve tried to, you know, speed it up as new models come out and things like that. Some clients are, you know, a couple months, and we build them in automation, and that’s, you know, all they need.
166 00:25:04.070 ⇒ 00:25:11.690 Samuel Roberts: so it varies quite a bit. You know, we’re definitely moving as we’ve grown to bigger and bigger clients, so I will kind of…
167 00:25:11.740 ⇒ 00:25:29.229 Samuel Roberts: bigger clients, longer engagements. As you can imagine, like, starting a consultancy, it’s hard to sell to big enterprise initially, so… but we’re getting to that point where I think engagements will last, you know, more months than they have, years, you know, working with the same people, which is nice. Okay.
168 00:25:29.230 ⇒ 00:25:30.830 Vimarsh Patel: Fair enough. That makes sense, yeah.
169 00:25:30.830 ⇒ 00:25:31.350 Samuel Roberts: Yeah.
170 00:25:31.790 ⇒ 00:25:38.580 Vimarsh Patel: That’s correct. Yeah, I think these are some other… no, these are on the top of the mind I had. Maybe, like, one or two, maybe at the end, I think so.
171 00:25:38.580 ⇒ 00:25:40.820 Samuel Roberts: I just wanted to… I like to leave time in the middle, because I don’t want to.
172 00:25:40.820 ⇒ 00:25:42.890 Vimarsh Patel: No, no, no, thank you, Nana, I appreciate it.
173 00:25:42.890 ⇒ 00:25:46.790 Samuel Roberts: So, actually, let’s… let’s talk a little bit about, your,
174 00:25:46.930 ⇒ 00:26:02.619 Samuel Roberts: I don’t want to say coding style, but your method. Like, obviously, there’s new coding agents and new tools and things like that, and so I’m curious how you typically go about building, you know, if you were starting a new project, and you had kind of a, you know, a rough
175 00:26:02.740 ⇒ 00:26:08.519 Samuel Roberts: requirements, and… and you could maybe start prototyping. How would you go about that? What is your.
176 00:26:08.520 ⇒ 00:26:27.999 Vimarsh Patel: Yes, sure. Yeah, absolutely. So, I think, like, if I’m building out something, I think my… my first… like, this was, like, before, obviously, the coding agents also came into play. Like, this was what I used to do. I should just go on GitHub and, like, search for what I’m building, because there’s so much, like, code available out there. Like, obviously, you don’t need to take the whole repository, but maybe just, like, an inspiration.
177 00:26:28.030 ⇒ 00:26:43.139 Vimarsh Patel: For, like, what actually is out there, because someone might have done it better than you, so why… why are you supposed to waste time into, like, you know, thinking about it? Like, obviously, if it’s not there, because it’s like… so, like, I heard this quote from Bill Gates, like, if someone has… if someone can do a task better, then you might as well
178 00:26:43.210 ⇒ 00:26:57.180 Vimarsh Patel: delegate it to them. So, like, if I do believe that there are better software engineers than me out there, so if someone has already, like, done something similar to my workflow, then might as well take inspiration and, like, not exactly copy, but at least learn from them.
179 00:26:57.180 ⇒ 00:26:58.300 Samuel Roberts: The problem is, yeah.
180 00:26:58.300 ⇒ 00:27:12.110 Vimarsh Patel: Yeah, exactly. So, I think that’s what I used to do, like, so if I’m building an e-commerce website, so, like, Amazon is obviously out there, but, like, you cannot get Amazon’s code. So the next best thing is, like, you go to GitHub and, like, see what other e-commerce
181 00:27:12.110 ⇒ 00:27:19.620 Vimarsh Patel: people have been. And especially, I think, the front-end part is, like, because I was not good at, like, the designing part.
182 00:27:19.620 ⇒ 00:27:31.099 Vimarsh Patel: So, I used to get the front-end code from, like, from GitHub, like a template or something, something sort of that thing, so I don’t have to waste time on that, and then the back-end, I can do it on my own, because I was really good at it. So…
183 00:27:31.190 ⇒ 00:27:34.260 Vimarsh Patel: I used to do that. Now I don’t need to worry about scraping, like, like.
184 00:27:34.360 ⇒ 00:27:38.159 Vimarsh Patel: like, the GitHub for front-end, like, I have Figma and, like, Coco.
185 00:27:38.890 ⇒ 00:27:46.579 Vimarsh Patel: Like, the agents for it. But yeah, so, like, first I would do that. Second would be, like, maybe make up a rough plan on a, like, like.
186 00:27:46.670 ⇒ 00:28:03.809 Vimarsh Patel: write it down with, like, what all the features I want, how the data would flow in the system, maybe, like, a small, like, a rough system design to understand, like, what flows from where, maybe, like, the different modules, or, like, the different layers of the system. That would be, like, my second approach, and then,
187 00:28:04.030 ⇒ 00:28:10.350 Vimarsh Patel: I think now, like, since coding agents are there, we can obviously use them to increase our speed and efficiency to code, like.
188 00:28:10.450 ⇒ 00:28:14.899 Vimarsh Patel: they can write thousands of codes in, like, a couple of seconds, so… Yeah. Yeah.
189 00:28:14.900 ⇒ 00:28:17.139 Samuel Roberts: What do you use in terms of coding agents?
190 00:28:17.140 ⇒ 00:28:25.970 Vimarsh Patel: So, I think till now, like, till January, I was using Codex, but then I recently shifted to Cloud Code. Okay. Because somehow, I think…
191 00:28:26.130 ⇒ 00:28:35.359 Vimarsh Patel: I, like, I have been, like, ever since ChatGPT came out, I was using ChatGPT since 2023, but I recently made the switch to Claude. I just…
192 00:28:35.360 ⇒ 00:28:49.029 Vimarsh Patel: like, somehow, I was not happy with the answers that I was getting out, so I tried… and one of my friends suggested that on technical terms and in terms of code, Claude was far superior, so I’m like, might as well give it a try, and I actually loved it, so I made the switch.
193 00:28:49.030 ⇒ 00:28:49.380 Samuel Roberts: Right.
194 00:28:49.380 ⇒ 00:28:52.929 Vimarsh Patel: So, yeah, and I think when it comes to using
195 00:28:53.100 ⇒ 00:29:17.429 Vimarsh Patel: the coding agents, I think your prompt and the context is very much important. I mean, in any use case, it is very important, but I think when it comes to code, it’s so much easier to get an output if you have all the requirements mentioned. And by requirements, I just don’t mean the features, but, like, all the exact workflows, like, how you want your system to be authenticated, what kind of authorization you want, what kind of rate limiting, like, so, like, all the things also, like, you
196 00:29:17.450 ⇒ 00:29:32.350 Vimarsh Patel: you need to mention it, like, you want to have locking capabilities as well, because otherwise then it will not do it. So I think, you also need to, like, it is also much easier to mention the stack that you want to use. So, so, like, it is a far better output when you
197 00:29:32.380 ⇒ 00:29:39.620 Vimarsh Patel: give all the requirements, all the small, like, the nitty-gritty details also to Claude, or Codex, Wahoo, whatever you’re using.
198 00:29:39.620 ⇒ 00:29:40.060 Samuel Roberts: Sure.
199 00:29:40.060 ⇒ 00:29:51.260 Vimarsh Patel: And I think even before that, I would also use their planning mode as well, to see what they have planned, because now that they also have it, they also have… they can be better planners sometimes. So…
200 00:29:51.450 ⇒ 00:30:09.569 Vimarsh Patel: use their planning mode, compare my plans, their plans, maybe make some tweaks, and then make the prompt again, and then see what output I get, test it out, and then go on iterating based on the testing, the errors that I keep getting, what exactly I wanted. This would be my, like, rough approach to building the prototype.
201 00:30:09.570 ⇒ 00:30:15.970 Samuel Roberts: Yeah, yeah. No, it’s really interesting. I’m always curious to hear how people are using these tools, because we’re all trying to figure out the best way to…
202 00:30:15.970 ⇒ 00:30:16.480 Vimarsh Patel: Yup, yeah.
203 00:30:16.480 ⇒ 00:30:24.480 Samuel Roberts: how much handwritten code is important, how much… how important is it to read the generated code and test it, and all this other stuff, so I’m always curious to hear how people are figuring this out.
204 00:30:24.480 ⇒ 00:30:24.960 Vimarsh Patel: Yeah, great.
205 00:30:24.960 ⇒ 00:30:42.679 Samuel Roberts: Great, okay. I think my last question, and hopefully we’ll have a minute or two if you have anything else, but we’re getting close, so I just wanted to finish up, but tell me about a time that you may have lost motivation on something you were working on, and how you handled that, and kind of what
206 00:30:43.080 ⇒ 00:30:44.740 Samuel Roberts: What you did about that.
207 00:30:45.850 ⇒ 00:30:47.439 Vimarsh Patel: lost motivation, so…
208 00:30:47.560 ⇒ 00:30:56.929 Vimarsh Patel: I think, yeah, so, like, like, till… until, like, this Christmas, like, 2025, December, I was working on a… on a SaaS tool for driving schools.
209 00:30:56.930 ⇒ 00:30:57.890 Samuel Roberts: Okay.
210 00:30:58.080 ⇒ 00:31:11.730 Vimarsh Patel: we were trying to, like, there are a couple of training schools in Canada that they wanted a software to be built for their internal, like, their, like, the scheduling, the payments, and, like, a… like, kind of like an ERP system, like a small, but, like, not that big.
211 00:31:11.730 ⇒ 00:31:22.299 Vimarsh Patel: And they were, like, a small school, so obviously they cannot afford to get those sales force and all those, so they wanted to build it in-house, like, and they… I… like, one of my friends is in Canada, and he somehow
212 00:31:22.340 ⇒ 00:31:29.059 Vimarsh Patel: got in touch, and he got the thing, so, like, I was helping out, and, like, the two of us were working on it. And it was, like,
213 00:31:29.530 ⇒ 00:31:36.289 Vimarsh Patel: We were doing good, we also gave in the demo and everything, and I think, October, November, we gave the initial demo, they were happy with it.
214 00:31:36.470 ⇒ 00:31:38.499 Vimarsh Patel: And then, in…
215 00:31:38.750 ⇒ 00:31:55.129 Vimarsh Patel: we got stuck somewhere, and then… and in December, we were supposed to roll it out. We gave it to them for testing, but they were not quite happy, and then they put us on a hold. So, they didn’t explicitly say no to us, but, like, at that time, it was a bit of a, like, like an…
216 00:31:55.130 ⇒ 00:32:00.119 Vimarsh Patel: bad kind of a feeling, like, it was, like, it was not… because, like, we put in so much effort over the past 6 months, or, like, maybe more.
217 00:32:00.120 ⇒ 00:32:00.609 Samuel Roberts: Right, right.
218 00:32:00.610 ⇒ 00:32:08.740 Vimarsh Patel: Because they kept giving us features that they wanted to implement, and then when we implemented, and then when we went out for
219 00:32:08.770 ⇒ 00:32:18.149 Vimarsh Patel: In the production, when it went to them for using, they were not happy. So I think what happened was, like, they opened a new branch, in… around the holidays.
220 00:32:18.150 ⇒ 00:32:30.509 Vimarsh Patel: And they wanted it to use the system there, because it was a new branch, so they thought that instead of moving all the data from the previous software to this new software, might as well use it on the new branch, since there’s no data for that.
221 00:32:30.510 ⇒ 00:32:34.090 Vimarsh Patel: And then once it… it works, It works out good.
222 00:32:34.190 ⇒ 00:32:45.470 Vimarsh Patel: we can implement it in other branches. So what happened was, like, we… so I think that since it was December, I think they didn’t get a lot of students that they were expecting, because of the holidays, or the weather, or…
223 00:32:45.580 ⇒ 00:32:53.199 Vimarsh Patel: Whatever it would have been. So, like, we didn’t get enough data on the software also, and they were not happy with how the software was working out for them.
224 00:32:53.240 ⇒ 00:33:01.350 Vimarsh Patel: So, I think it was a bit of a disappointment. But I think then, come January, it did come out, they still wanted to go ahead, but they wanted…
225 00:33:01.370 ⇒ 00:33:13.630 Vimarsh Patel: They said that maybe we can let go of the new branch, because it was not working out for them also. And we are still working on it, like, it’s kind of a side project thing that we are doing, so it takes a little bit of time, but it’s like.
226 00:33:13.630 ⇒ 00:33:22.680 Vimarsh Patel: And then we sat down, we realized that building a normal SaaS tune for a training school is also a bit of an overkill, because anyone can do it. We need to give something different as well. Maybe, so…
227 00:33:22.680 ⇒ 00:33:23.170 Samuel Roberts: Alright, yeah.
228 00:33:23.170 ⇒ 00:33:29.000 Vimarsh Patel: We, again, talked with them, we also thought of, like, we are thinking of putting in a small AI agent for students.
229 00:33:29.020 ⇒ 00:33:44.930 Vimarsh Patel: That can help with their scheduling and rescheduling and all that, so they don’t have to manually use the software, they can just chat with the agent, and it can book the schedules for them. And even, like, they… like, some of the driving school software is also given, like, tests for the students, like, the return test practice and everything.
230 00:33:44.930 ⇒ 00:33:45.430 Samuel Roberts: Sure.
231 00:33:45.430 ⇒ 00:33:52.869 Vimarsh Patel: So, we are trying to do that as well, so maybe it can, like, the software can be as possible for the driving school.
232 00:33:52.980 ⇒ 00:34:09.209 Vimarsh Patel: So I think that’s how I think we came up, like, we realized that maybe it was not… like, obviously the software needed improvement, but I think it was not just us that was lagging. Like, the new brands also didn’t get any, like, enough people coming in, so the admin department couldn’t use the software as well. So I think that it was, like, a.
233 00:34:09.219 ⇒ 00:34:09.889 Samuel Roberts: Right.
234 00:34:10.090 ⇒ 00:34:12.949 Vimarsh Patel: The, like, a to-and-flow relationship, so… Hmm.
235 00:34:13.260 ⇒ 00:34:13.880 Vimarsh Patel: But, yeah.
236 00:34:14.530 ⇒ 00:34:24.460 Samuel Roberts: Great, no, thank you, that’s wonderful context, yeah. Okay, so we’re just about at time. I’m not… I don’t have a hard stop, but I want to just, again, if you have any other questions or anything.
237 00:34:24.590 ⇒ 00:34:32.980 Vimarsh Patel: I don’t think I have any other questions, but, like, it was amazing talking to you as well, but, like, I would like to know, like, if the second route comes up, then… That’s it.
238 00:34:32.980 ⇒ 00:34:35.939 Samuel Roberts: That’s exactly what I was gonna… that was the last thing I was gonna get to, so…
239 00:34:35.940 ⇒ 00:34:37.069 Vimarsh Patel: Okay, perfect, perfect.
240 00:34:37.070 ⇒ 00:34:55.000 Samuel Roberts: It’s kind of a three-step process, so this is the first interview. I’ll bring this back to the team, and if you pass that round, you would have a second, maybe slightly more technical, role-focused interview with another engineer on the team. And then, if that goes well, there’s a panel interview.
241 00:34:55.000 ⇒ 00:34:58.330 Samuel Roberts: That would be about a tech challenge that we would
242 00:34:58.330 ⇒ 00:35:03.169 Samuel Roberts: give you, to build a little tool. And, you know, it’s pretty…
243 00:35:03.310 ⇒ 00:35:26.260 Samuel Roberts: you know, it’s a defined problem, but it’s… you have the freedom to figure out how to do it, and however you want to build it, and then, basically present it to us, and we’ll look at the code, and we’ll talk about it, and, see how… see how it all goes. So that’s the whole… that’s the process, and so it’s like, pass this one, go to the second, pass that one, go to the third, and then that would be the decision. So we try to go relatively quickly. We don’t want to drag things out too much, but obviously scheduling is always a…
244 00:35:26.260 ⇒ 00:35:26.700 Samuel Roberts: An issue.
245 00:35:26.700 ⇒ 00:35:27.500 Vimarsh Patel: Yeah, I know.
246 00:35:27.500 ⇒ 00:35:32.580 Samuel Roberts: But yeah, I mean, so look forward to hearing one way or another, hopefully relatively soon, and then…
247 00:35:32.580 ⇒ 00:35:33.330 Vimarsh Patel: So the second round.
248 00:35:33.330 ⇒ 00:35:34.569 Samuel Roberts: My next one. Yeah.
249 00:35:34.570 ⇒ 00:35:38.090 Vimarsh Patel: Will it be, like, a live coding, or, like, just a discussion like this?
250 00:35:38.090 ⇒ 00:35:41.780 Samuel Roberts: A session like this, but more technical. So this one is a little more, you know.
251 00:35:41.780 ⇒ 00:35:42.240 Vimarsh Patel: Yeah. I talked.
252 00:35:42.240 ⇒ 00:35:44.980 Samuel Roberts: a little more general and a little more culture-fit stuff, and then…
253 00:35:44.980 ⇒ 00:35:45.340 Vimarsh Patel: Mix it.
254 00:35:45.340 ⇒ 00:35:49.700 Samuel Roberts: That one, the engineer will dive in a little more about Okay.
255 00:35:49.700 ⇒ 00:35:51.249 Vimarsh Patel: Perfect. No, no, yeah, got it, got it.
256 00:35:51.470 ⇒ 00:35:53.560 Samuel Roberts: But not, yeah, don’t worry about coding live, then.
257 00:35:53.560 ⇒ 00:35:53.900 Vimarsh Patel: Okay.
258 00:35:53.900 ⇒ 00:35:54.640 Samuel Roberts: And the.
259 00:35:54.640 ⇒ 00:35:56.469 Vimarsh Patel: I just want to be prepared.
260 00:35:56.470 ⇒ 00:36:02.219 Samuel Roberts: I completely understand. And then the other one, like I said, it will talk about the code, but it won’t be live coding. You’ll have already.
261 00:36:02.220 ⇒ 00:36:02.590 Vimarsh Patel: I don’t.
262 00:36:02.590 ⇒ 00:36:03.360 Samuel Roberts: So, okay.
263 00:36:03.360 ⇒ 00:36:04.190 Vimarsh Patel: Got it, makes sense.
264 00:36:04.190 ⇒ 00:36:04.570 Samuel Roberts: Alright, great.
265 00:36:04.570 ⇒ 00:36:07.390 Vimarsh Patel: That sounds wonderful. Yeah, I’m hoping to hear from you.
266 00:36:07.390 ⇒ 00:36:10.329 Samuel Roberts: Yeah, thank you. Hopefully you’ll hear from us very soon. Alright, thank you so much.
267 00:36:10.330 ⇒ 00:36:11.940 Vimarsh Patel: Thank you. Thank you so much, have a great day.
268 00:36:12.330 ⇒ 00:36:13.390 Samuel Roberts: You too. Bye-bye.