Meeting Title: Data Engineer Interview (Ujval Kamath) Date: 2026-01-23 Meeting participants: Awaish Kumar, Ujval Kamath
WEBVTT
1 00:00:39.400 ⇒ 00:00:40.200 Awaish Kumar: Hi.
2 00:00:40.870 ⇒ 00:00:41.540 Ujval Kamath: Hi.
3 00:00:43.650 ⇒ 00:00:44.740 Awaish Kumar: How you doing?
4 00:00:45.410 ⇒ 00:00:46.560 Ujval Kamath: I’m doing well, how are you?
5 00:00:46.790 ⇒ 00:00:47.919 Awaish Kumar: I’m good as well.
6 00:00:48.150 ⇒ 00:00:51.380 Awaish Kumar: So, it’s called, like, Ujwan, right?
7 00:00:51.910 ⇒ 00:00:52.510 Ujval Kamath: Yeah.
8 00:00:52.700 ⇒ 00:00:55.490 Ujval Kamath: And, yours, I assume, is a wish.
9 00:00:55.700 ⇒ 00:01:03.490 Awaish Kumar: Yep, okay. Thank you for taking the time for this interview. So, like, I’m Arish Kumar.
10 00:01:03.590 ⇒ 00:01:10.039 Awaish Kumar: I work, At Brainforge as a… as a lead data engineer.
11 00:01:11.940 ⇒ 00:01:15.370 Awaish Kumar: I have been, basically, Oh, this…
12 00:01:15.840 ⇒ 00:01:20.000 Awaish Kumar: Like, I’ve been working as a data engineer for more than 9 years now.
13 00:01:20.300 ⇒ 00:01:21.220 Awaish Kumar: Oh, boy.
14 00:01:21.940 ⇒ 00:01:27.779 Awaish Kumar: So yeah, that’s basically it, and for this interview, we will be just mostly… we’ll be learning about your
15 00:01:28.310 ⇒ 00:01:34.950 Awaish Kumar: Past experiences, projects, and maybe deep dive into technical,
16 00:01:35.330 ⇒ 00:01:41.240 Awaish Kumar: Details of each, like, individual projects, and how did you contribute it on those, that’s all.
17 00:01:41.640 ⇒ 00:01:42.580 Awaish Kumar: Sure.
18 00:01:43.200 ⇒ 00:01:46.430 Awaish Kumar: Okay, yeah, you can start with your introduction.
19 00:01:47.420 ⇒ 00:02:00.750 Ujval Kamath: Sure. So, yeah, so I’m Ujwal. I’ve been working as, in data science roles for about, a little… about a decade now. I was briefly a data science manager, but mostly it’s been individual contributor roles.
20 00:02:01.010 ⇒ 00:02:07.999 Ujval Kamath: I’ve had, like, a mix of, like, external-facing, so customer-facing roles, and internal-facing roles.
21 00:02:08.120 ⇒ 00:02:09.889 Ujval Kamath: All of my…
22 00:02:10.160 ⇒ 00:02:26.620 Ujval Kamath: individual… all of my jobs, typically, I… I sort of had the similar… like, I had similar expectations, which is that, I’ll eat… I… I’ll work with some customer or some… either a customer or a stakeholder who has some sort of… they have some data, they have…
23 00:02:27.020 ⇒ 00:02:40.059 Ujval Kamath: a business problem, and they’re just trying to understand, like, oh, I want to use predictive analytics or data science or something like that. And it’s really my job to sort of help them go from, like, just idea, like, concept.
24 00:02:40.060 ⇒ 00:02:52.479 Ujval Kamath: to evaluate, like, you know, what’s the business feasibility? What’s sort of the cost-benefit? Is this even realistic? Do we have the data from this? And assuming, like, that’s done, then I would typically build out something from
25 00:02:52.750 ⇒ 00:02:56.229 Ujval Kamath: That idea to something that would go into production.
26 00:02:56.380 ⇒ 00:03:06.370 Ujval Kamath: In terms of, technical skills, I am pretty comfortable with, I say, and I’ve worked primarily in manufacturing and healthcare.
27 00:03:06.530 ⇒ 00:03:08.640 Ujval Kamath: In terms of…
28 00:03:08.870 ⇒ 00:03:23.080 Ujval Kamath: Technical skills, I think I’m comfortable with, sort of, the data science stack. I think since you’re a data engineer, I’ll just maybe qualify this, and then I have done a lot of work with building data pipelines and everything. I don’t have that much experience on the data warehousing side of things.
29 00:03:23.230 ⇒ 00:03:27.309 Ujval Kamath: So typically, If there’s, like,
30 00:03:27.690 ⇒ 00:03:36.039 Ujval Kamath: some, like… let’s say somebody has to build, like, a data warehouse, that’s typically done by a data engineer, right? So,
31 00:03:36.400 ⇒ 00:03:51.180 Ujval Kamath: So I would typically take something that’s in an existing table. It’s not always been the case, I have worked with… I have actually built up stuff, but typically, like, the… the data, it may be a mess, but it’ll be in a data warehouse, and I’ll sort of build the data pipeline down from there.
32 00:03:52.470 ⇒ 00:03:57.040 Ujval Kamath: I do have some experience with data warehousing, but, like, I don’t want to say I have a lot, it’s quite limited.
33 00:03:57.240 ⇒ 00:04:01.460 Ujval Kamath: So I’m comfortable, like, in a while. I’ve worked on on-prem cloud stuff.
34 00:04:01.920 ⇒ 00:04:03.349 Ujval Kamath: I’m comfortable with.
35 00:04:04.230 ⇒ 00:04:06.050 Ujval Kamath: Different, tool stacks.
36 00:04:06.330 ⇒ 00:04:09.739 Ujval Kamath: So… Well, that gives a quick background.
37 00:04:09.930 ⇒ 00:04:10.250 Awaish Kumar: Okay.
38 00:04:10.250 ⇒ 00:04:12.349 Ujval Kamath: We can talk about some use cases if you’re interested, but…
39 00:04:12.910 ⇒ 00:04:17.260 Awaish Kumar: Yeah, so… I… what I would like to understand, now that
40 00:04:17.880 ⇒ 00:04:22.070 Awaish Kumar: Yeah, given your introduction, mostly it’s, like, it’s…
41 00:04:22.280 ⇒ 00:04:30.919 Awaish Kumar: The front-facing, like, the client-facing data scientist, where you are mostly working Showcasing the…
42 00:04:31.240 ⇒ 00:04:34.830 Awaish Kumar: Maybe helping answer some of the business-related questions, or.
43 00:04:34.830 ⇒ 00:04:35.210 Ujval Kamath: Alright.
44 00:04:35.210 ⇒ 00:04:46.179 Awaish Kumar: investigations. So I would like to answer, like, I would like to understand, like, how that… what is the process, like, of doing that? Like.
45 00:04:46.690 ⇒ 00:04:54.140 Awaish Kumar: For example, right now, I have a customer who has the data living in their data source.
46 00:04:54.840 ⇒ 00:04:58.700 Awaish Kumar: It’s an e-commerce, like, it’s some kind of retail data, right?
47 00:04:59.330 ⇒ 00:04:59.900 Ujval Kamath: Right.
48 00:05:00.440 ⇒ 00:05:05.990 Awaish Kumar: So how… what would we do? Like, data is sitting there, client don’t know what to do with it.
49 00:05:06.380 ⇒ 00:05:08.519 Awaish Kumar: But there is an enormous amount of data.
50 00:05:08.840 ⇒ 00:05:11.139 Awaish Kumar: So, what would be your steps to…
51 00:05:11.240 ⇒ 00:05:15.370 Awaish Kumar: coming to a point where you can actually support the business.
52 00:05:15.730 ⇒ 00:05:16.940 Awaish Kumar: Like, Clan knows… Sure.
53 00:05:17.580 ⇒ 00:05:21.440 Awaish Kumar: We have data, we can do something about it, but what we can do
54 00:05:21.820 ⇒ 00:05:24.430 Awaish Kumar: They don’t… they don’t know. They need help with that.
55 00:05:25.650 ⇒ 00:05:33.179 Ujval Kamath: Sure. So, I mean, I think the first thing I typically do is try to understand, maybe sort of where they are in terms of their own maturity.
56 00:05:33.390 ⇒ 00:05:38.750 Ujval Kamath: You know, sometimes you have customers who say, oh, I have data, but they don’t even really know what it is.
57 00:05:38.850 ⇒ 00:05:45.659 Ujval Kamath: So, you know, they’re really on the business side, so they… maybe they’re in operations or something, but they don’t really… they’re like, oh, we have data.
58 00:05:45.930 ⇒ 00:05:48.790 Ujval Kamath: But they… they can’t really describe it well. Yeah.
59 00:05:48.790 ⇒ 00:06:07.380 Awaish Kumar: That is here. And my question was, what I’m asking is that customer has an operational knowledge, they are million dollar, millions of dollars of business, right? They are running it, but there’s a lot of data sitting there, which is, like, which is not utilized. So how?
60 00:06:07.380 ⇒ 00:06:08.080 Ujval Kamath: Yep.
61 00:06:08.080 ⇒ 00:06:15.249 Awaish Kumar: We are going to suggest some strategies. How can you use this data to answer some of your operational stuff?
62 00:06:16.640 ⇒ 00:06:24.249 Ujval Kamath: Sure, so, I mean, I think I would do a couple of things. One is, of course, you know, they have a specific industry.
63 00:06:24.450 ⇒ 00:06:39.920 Ujval Kamath: They probably have specific use cases, depending on which industry they’re in, which part of the business they’re in. Certainly, I mean, before I would even have a meeting with them, I would probably do my own research, and now with, you know, chat GPT LLMs, that’s actually much easier than it used to be.
64 00:06:39.950 ⇒ 00:06:53.960 Ujval Kamath: is just to give me that sort of lingo and, like, the know-how about what some possible use cases would be. Then I think I… what I typically do is I will just go to them and I will try to understand, like, what are they… what are some of the KPIs they have?
65 00:06:54.140 ⇒ 00:07:02.530 Ujval Kamath: What are some of the pain points they have? And then maybe, what are some of, like, processes that they’re doing now that are really, really manual?
66 00:07:02.840 ⇒ 00:07:05.230 Ujval Kamath: That could be automated.
67 00:07:05.580 ⇒ 00:07:21.039 Ujval Kamath: Like, you know, like, maybe they’re doing something where they’re like, oh, right now, every month, we have some sort of thing that takes two full days, and we’re doing it in Excel, and we’re copying and pasting, and I’m sending files to this person, that person, this kind of a mess. And I think to some degree, like, this is…
68 00:07:21.340 ⇒ 00:07:24.070 Ujval Kamath: Use that, sort of, to identify, like.
69 00:07:24.350 ⇒ 00:07:43.690 Ujval Kamath: where they really, like, need something to help, right? So, you know, I think in data science, it’s very easy to, like, pitch, like, a very, very complicated, fancy solution, but that’s not their pain point, right? For them, they’re, like, it’s much more basic. And I think part of that is also to help me evaluate, like, do they really need a data scientist, or do they need, like, a data engineer?
70 00:07:43.760 ⇒ 00:07:45.049 Ujval Kamath: So, I mean, I’ve…
71 00:07:45.050 ⇒ 00:07:45.520 Awaish Kumar: I’m…
72 00:07:45.520 ⇒ 00:07:46.240 Ujval Kamath: Yeah.
73 00:07:46.240 ⇒ 00:07:54.939 Awaish Kumar: In our case, like, what I’m describing, we have data engineers there working and building documented solutions.
74 00:07:55.180 ⇒ 00:08:05.809 Awaish Kumar: Now, I’m just looking at it from data scientist’s point of view. Right. Maybe I can call, like, a strategist person who can come on board, look at the data.
75 00:08:06.070 ⇒ 00:08:12.949 Awaish Kumar: And then what would be the process of coming to a point where you can actually convince the… the… the…
76 00:08:13.450 ⇒ 00:08:17.380 Awaish Kumar: Stakeholder that, okay, you have been doing this
77 00:08:17.540 ⇒ 00:08:22.519 Awaish Kumar: And, like, obviously, they are making some decisions, somehow, right now.
78 00:08:22.540 ⇒ 00:08:33.879 Awaish Kumar: But here are the improved ways of doing that. Here’s how you can optimize your decisions. Like, it’s not more about automation, like, optimizations on terms of,
79 00:08:33.880 ⇒ 00:08:49.809 Awaish Kumar: automation. I’m more talking about, like, if I’m making a guess, like, what would be my revenue in the next month, right? What could I… that needs to be, like, a guided guess, not just, like, something I
80 00:08:50.520 ⇒ 00:08:51.470 Awaish Kumar: I agree.
81 00:08:51.470 ⇒ 00:08:52.810 Ujval Kamath: You mean, like, a forecast?
82 00:08:54.520 ⇒ 00:09:02.370 Awaish Kumar: Yeah, that’s, like… I don’t mean exactly a forecast, like, my… my point is more like, this was just an example.
83 00:09:02.570 ⇒ 00:09:11.640 Awaish Kumar: like, there are a lot of questions similar to that, like, for example, I want to ensure my inventory of my products is,
84 00:09:12.050 ⇒ 00:09:24.090 Awaish Kumar: is optimized, like, I don’t want to send over, like, inventory where, like, there are not much sales, I don’t want to send it where it is underutilized, so…
85 00:09:24.430 ⇒ 00:09:30.469 Awaish Kumar: Right now, I’m just making a decision out of blue, like, okay, I think this area would be…
86 00:09:30.760 ⇒ 00:09:47.099 Awaish Kumar: there will be a lot of sales, right? Like, I think kind of things. So, how would you go for identifying things where you hear, like, I think, from I think, to, like, the data is supporting these, basically, the decisions?
87 00:09:47.920 ⇒ 00:09:56.280 Ujval Kamath: Sure, I mean, so typically… okay, so let’s just assume I’ve talked to the customer, I’ve talked to the customer, I’ve had some conversations with them, they’ve identified.
88 00:09:56.460 ⇒ 00:09:57.510 Awaish Kumar: It’s more like…
89 00:09:57.510 ⇒ 00:09:58.110 Ujval Kamath: some…
90 00:09:58.250 ⇒ 00:10:02.369 Awaish Kumar: From… you have the data, how… now, how would you start your convo, like.
91 00:10:02.610 ⇒ 00:10:13.799 Awaish Kumar: when… what do you do before your conversation? How… what would you do when you meet with the client? And then, what would be the next steps? Like, I don’t need exact details, I just need, like.
92 00:10:13.960 ⇒ 00:10:16.510 Awaish Kumar: A process, a roadmap of how do you…
93 00:10:16.650 ⇒ 00:10:19.159 Awaish Kumar: Follow, like, few steps to reach there.
94 00:10:20.390 ⇒ 00:10:26.699 Ujval Kamath: I mean, so I… I’ll be honest, like, I… I’ve… in my years as a data scientist, my experience is, like.
95 00:10:27.030 ⇒ 00:10:35.699 Ujval Kamath: just having some data and doing some things and then going to the customer, it never works. Because the truth is that, you know, 10 people can look at the data
96 00:10:35.860 ⇒ 00:10:40.259 Ujval Kamath: And they can say, have 10 different ideas of what’s useful from it.
97 00:10:40.470 ⇒ 00:10:44.769 Ujval Kamath: So, at least for me, I think if I had that data, I might…
98 00:10:44.970 ⇒ 00:10:55.099 Ujval Kamath: I don’t think I would dig into it that much, I would just try to understand it maybe from a very, very high level. So, okay, let’s just say that we have orders data, we have,
99 00:10:55.570 ⇒ 00:11:14.399 Ujval Kamath: Okay, like, okay, we have some order tracking, we have, you know, I have some general context for what the data is. I don’t think I would put a lot of time initially into investing, like, oh, what columns we have and all that stuff. I would just, like, get a very, very high level of the data. I would then go to the customer, and then, like, if I’m talking to them about pain points.
100 00:11:16.010 ⇒ 00:11:20.799 Ujval Kamath: I would try to say, like, okay, you know, you said that you have issues with operations. I know we have
101 00:11:20.930 ⇒ 00:11:28.990 Ujval Kamath: You said you have some issues with, like, orders. I know we have data for that, and I think that would give me an opportunity to maybe talk about that, and at least say.
102 00:11:29.160 ⇒ 00:11:43.670 Ujval Kamath: okay, you listed these 10 things that are pain points. Five of them, we don’t even have data, so I, you know, I can immediately say we can’t focus on those. And I could maybe say that, okay, these 5 things, we do have data, maybe we can focus on them, and then I would maybe focus on business value.
103 00:11:43.750 ⇒ 00:11:54.560 Ujval Kamath: So, of those five, they might say that, oh, two of them are nice to have, but, you know, we don’t care. And then those three, and then based on those three, I think I would then go back into the data and spend more time
104 00:11:54.660 ⇒ 00:12:00.709 Ujval Kamath: trying to see if there’s anything related to what they’re trying to do. I think that’s typically how I would do it.
105 00:12:00.910 ⇒ 00:12:03.670 Awaish Kumar: Okay, Nick, what are your ways for deliverables?
106 00:12:04.710 ⇒ 00:12:07.600 Awaish Kumar: And how you deliver your work to the client.
107 00:12:08.910 ⇒ 00:12:15.729 Ujval Kamath: So typically, everything I… in terms of what I delivered was some sort of pro… Some sort of…
108 00:12:16.110 ⇒ 00:12:31.470 Ujval Kamath: machine learning or project… thing that was running in production, like, was running continuously in production. So I can give you an example. I worked in, or predicting issues with the machine, like, with medical devices, and typically what that means is that
109 00:12:31.540 ⇒ 00:12:48.500 Ujval Kamath: you identify specific issues that they want, like, sort of, you know, this service comp- service side of the operations side wants to capture, develop a model to predict when those issues are going to occur, and then those models were, like, written in Python and deployed in some sort of
110 00:12:48.660 ⇒ 00:12:55.519 Ujval Kamath: infrastructure, like, you know, it could be Databricks, it could be an AWS, and it’s continuously running with, like, new data that, like.
111 00:12:55.820 ⇒ 00:13:01.780 Ujval Kamath: Sends out a prediction. That prediction, like, goes into, like, another system which, like, consumes that.
112 00:13:01.950 ⇒ 00:13:03.379 Awaish Kumar: And I can give more…
113 00:13:03.810 ⇒ 00:13:08.900 Awaish Kumar: Yeah, what are… basically… what was your role? Like, were you also deploying the models, or…
114 00:13:09.170 ⇒ 00:13:17.669 Awaish Kumar: You were kind of a… only the data scientist who’s basically running the research and… and… and the, ex…
115 00:13:18.010 ⇒ 00:13:32.210 Awaish Kumar: Like, just the… and coming up with some optimal solutions in terms of a model or whatever, or were you actually deploying it in a cloud system, like, as an ML engineer? How… what was your scope of your work?
116 00:13:32.920 ⇒ 00:13:37.489 Ujval Kamath: So my scope was everything, like, after it came out of the data warehouse, that was MySculpt.
117 00:13:37.790 ⇒ 00:13:41.399 Ujval Kamath: So, I… I would be the one who started from
118 00:13:41.540 ⇒ 00:13:50.279 Ujval Kamath: You know, the dirty, the messy data, understand, do the data engineering, build a model, and then deploy it into, actually deploy it into production.
119 00:13:51.110 ⇒ 00:13:55.170 Awaish Kumar: Okay, so where, like, what were you using to deploy it?
120 00:13:56.130 ⇒ 00:14:05.060 Ujval Kamath: So I’ve used different things, so, depending on the job, so in… in… when I was in Siemens Healthcare, it was all on Databricks.
121 00:14:05.180 ⇒ 00:14:24.449 Ujval Kamath: So we had all the data in Delta Lake tables, we typically write, you know, PySpark scripts or Python scripts, in Databricks itself, and then the infrastructure was in Databricks as well, so we could deploy it, and essentially, there was just sort of a production system, and then our development system.
122 00:14:24.450 ⇒ 00:14:32.639 Awaish Kumar: I wanted to understand more, like, how were you deploying it? Like, were you deploying it as a service, or were you deploying… which is running,
123 00:14:32.770 ⇒ 00:14:35.639 Awaish Kumar: like, on request, like, somebody called.
124 00:14:35.640 ⇒ 00:14:37.740 Ujval Kamath: Alright, sure, so… So…
125 00:14:37.740 ⇒ 00:14:44.199 Awaish Kumar: Was it just a bad job doing predictions in the backend? How was it, like, actually working?
126 00:14:44.850 ⇒ 00:14:54.639 Ujval Kamath: So, I’ve done it two ways. So, when I was in… in Siemens Healthcare, I was… it was a… it was in batch, it ran 6… like, 6 times a day. Yeah, every 4 hours.
127 00:14:54.850 ⇒ 00:15:06.149 Ujval Kamath: And typically, it’s… I can talk more about the infrastructure, but there was, like, a central data pool that happened in one script, then each model would,
128 00:15:06.410 ⇒ 00:15:11.939 Ujval Kamath: filter that data down, so we weren’t doing multiple pulls in the data warehouse. And then…
129 00:15:12.560 ⇒ 00:15:17.299 Ujval Kamath: the… in another job, I actually was… it was exposed as an API.
130 00:15:17.460 ⇒ 00:15:21.380 Ujval Kamath: So the model was expo… it was running in an EC2 container.
131 00:15:21.560 ⇒ 00:15:23.040 Ujval Kamath: And then…
132 00:15:23.180 ⇒ 00:15:30.679 Ujval Kamath: somebody could essentially just make an API call to it. They, like, send the data, and then they’d get a response. So it was running 24 hours a day.
133 00:15:31.210 ⇒ 00:15:35.519 Awaish Kumar: So, like, you didn’t use any services like Cloud Functions or anything?
134 00:15:35.900 ⇒ 00:15:37.000 Awaish Kumar: Lambda functions.
135 00:15:37.000 ⇒ 00:15:41.890 Ujval Kamath: I mean, I’ve used Lambda, but typically, I’ve never deployed a model in Lambda.
136 00:15:42.230 ⇒ 00:15:42.810 Awaish Kumar: Okay.
137 00:15:42.810 ⇒ 00:15:49.449 Ujval Kamath: It’s, it’s almost, typically they’re a little… they, the, scoring time is sometimes
138 00:15:50.250 ⇒ 00:15:57.990 Ujval Kamath: The reason we didn’t use Lambda in the… In the former cases, the…
139 00:15:58.950 ⇒ 00:16:17.229 Ujval Kamath: the load was a little unpredictable, so the safest thing… and I didn’t make this decision, it was made by, like, our… there was an infrastructure… like, AWS infrastructure team. They said that they don’t want people deploying models in Lambda, they want just, like, have a small EC2 instance, so I don’t necessarily think that’s the best way to do it, but that’s…
140 00:16:17.350 ⇒ 00:16:20.940 Ujval Kamath: the way that it was decided to be done.
141 00:16:21.260 ⇒ 00:16:22.270 Awaish Kumar: Okay.
142 00:16:22.270 ⇒ 00:16:24.510 Ujval Kamath: Databricks, yeah, I didn’t use Cloud Functions.
143 00:16:25.190 ⇒ 00:16:31.519 Awaish Kumar: Okay, so… Like, do you know about this role? Like, how…
144 00:16:31.930 ⇒ 00:16:35.029 Awaish Kumar: what would be… what Drew will be doing, and, like.
145 00:16:35.600 ⇒ 00:16:38.460 Awaish Kumar: And are you comfortable with that?
146 00:16:39.940 ⇒ 00:16:42.840 Ujval Kamath: Yeah, I mean, I think based on what I’ve said is that typically.
147 00:16:43.020 ⇒ 00:16:46.679 Ujval Kamath: I think you would work with the systems the customer has, right?
148 00:16:47.790 ⇒ 00:16:59.179 Ujval Kamath: Or you’re using… so in terms of… I understand what the data science would be, but my understanding is that there isn’t, like, oh, you are on AWS only, AWS. My understanding is that you typically flex between the customers.
149 00:17:01.240 ⇒ 00:17:02.060 Awaish Kumar: I’m not sure.
150 00:17:02.060 ⇒ 00:17:02.700 Ujval Kamath: literature.
151 00:17:03.000 ⇒ 00:17:09.419 Awaish Kumar: like, for this role, like, Ruth will be kind of one of our customer-facing data engineers.
152 00:17:09.420 ⇒ 00:17:10.010 Ujval Kamath: Yeah.
153 00:17:10.329 ⇒ 00:17:10.909 Awaish Kumar: Right.
154 00:17:13.209 ⇒ 00:17:14.149 Awaish Kumar: Or…
155 00:17:14.150 ⇒ 00:17:15.720 Ujval Kamath: Interfacing data engineer?
156 00:17:16.030 ⇒ 00:17:19.009 Awaish Kumar: Yeah, like, either… like, we have a few…
157 00:17:19.500 ⇒ 00:17:24.180 Awaish Kumar: categories here, like data engineer, or you could be a data analyst.
158 00:17:24.300 ⇒ 00:17:37.559 Awaish Kumar: Or you can be a scientist, but in this category of scientist, we, we’re not, like, the most of the work, you know, is making the strategy, or, like, the… doing the competitive analysis, or things like that.
159 00:17:38.080 ⇒ 00:17:45.970 Awaish Kumar: And then, obviously, you can support with, okay, we need some forecasting, we can use ML for that, right?
160 00:17:45.970 ⇒ 00:17:46.710 Ujval Kamath: Right.
161 00:17:46.710 ⇒ 00:17:52.989 Awaish Kumar: But it depends on the use cases. It’s not always that you will be working on
162 00:17:53.150 ⇒ 00:17:54.799 Awaish Kumar: Just building the models.
163 00:17:54.900 ⇒ 00:18:06.980 Awaish Kumar: Right? It… it will be sometimes, you know, you will be work… can… you will be working on, maybe, classification work, you know, building a classification model, or building a forecasting model, or something like that, or maybe…
164 00:18:06.980 ⇒ 00:18:15.240 Awaish Kumar: We also have a team of AI engineers, which are doing, kind of, some work with LLMs and all of the agents and things like that.
165 00:18:15.290 ⇒ 00:18:23.999 Awaish Kumar: But, having said that, what I’m assuming, the majority of work is that helping clients
166 00:18:24.110 ⇒ 00:18:32.619 Awaish Kumar: understand the business impact of the things, right? As I said, like, one of the examples
167 00:18:32.630 ⇒ 00:18:44.789 Awaish Kumar: I gave, like, we have data, they want to do analysis, right? They don’t know right yet what to do. So we can’t start, like, as a step one, that, okay, let’s just use some ML, right?
168 00:18:44.850 ⇒ 00:18:47.690 Awaish Kumar: That’s… that’s not… Right. It’s not going to work.
169 00:18:47.690 ⇒ 00:18:48.240 Ujval Kamath: It doesn’t work.
170 00:18:48.240 ⇒ 00:18:59.399 Awaish Kumar: We are going to come up with some analysis, we are going to show something, like, with this data, what we can do in terms of helping you make some decisions on your operationals.
171 00:18:59.910 ⇒ 00:19:01.570 Awaish Kumar: And your operations.
172 00:19:01.850 ⇒ 00:19:12.709 Awaish Kumar: After that, like, there could be opportunities that, okay, you can… and that’s where you are going to identify, because clients, they don’t have a data team, right?
173 00:19:12.850 ⇒ 00:19:21.459 Awaish Kumar: they, they, like, they joined, like, they, signed a contract with Brainforge is because they don’t have a team of data
174 00:19:22.050 ⇒ 00:19:27.320 Awaish Kumar: Scientists and engineers. So, like, you are the one who is going to, for example, if you are
175 00:19:27.830 ⇒ 00:19:31.330 Awaish Kumar: hired as a data scientist, who will be the one, like, you come up with
176 00:19:31.450 ⇒ 00:19:42.130 Awaish Kumar: some ideas, like, okay, this is one of our customers, we are doing this work for them, and here is how we can use ML to
177 00:19:42.730 ⇒ 00:19:50.100 Awaish Kumar: Help them, like, improve, optimize, whatever, like, add new features to this.
178 00:19:50.100 ⇒ 00:19:50.470 Ujval Kamath: Sure.
179 00:19:50.470 ⇒ 00:19:51.880 Awaish Kumar: Then,
180 00:19:52.200 ⇒ 00:20:00.570 Awaish Kumar: That’s how we are going to propose to client, right? That’s how… what we can do for you, and they are going to approve it, right? That’s how it’s going to work.
181 00:20:00.690 ⇒ 00:20:05.790 Awaish Kumar: So, yeah, so it’s… It will be analysis, or a lot of analysis.
182 00:20:06.200 ⇒ 00:20:23.979 Awaish Kumar: And obviously, all the analysis is customer-facing. You may be building the presentation, or you may be building some dashboard in Tableau, Power BI, whatever tool it is, and basically showcasing the client, like, what is the work?
183 00:20:25.360 ⇒ 00:20:43.579 Ujval Kamath: Sure, I mean, I’m comfortable with that. I have done it quite a bit in the past. I’ve… so I… it’s not something that… I’ve never been in a job where I… like, I’m so focused on the ML part that I have to do that all the time. Like, I’ve always been in part of, like, the data team, even if I’m at a large company, the data team where I am tends to be pretty small.
184 00:20:43.750 ⇒ 00:20:54.529 Ujval Kamath: So we don’t have that luxury of, like, Google, you know, where it’s like, oh, your job is just to tweak the ML algorithm by 1%. I’m very comfortable with, sort of, the business side of things as well, so I’m not…
185 00:20:54.530 ⇒ 00:21:03.750 Awaish Kumar: Now, apart from that, yeah, we, we, yeah, in Brainforce, you might be working on more than one client at a time.
186 00:21:04.040 ⇒ 00:21:13.939 Awaish Kumar: We… yeah, that’s… and that’s also possible. And yeah, that… having said that, like, that was just… I’m… I was setting the context to,
187 00:21:14.060 ⇒ 00:21:16.619 Awaish Kumar: To align on what we are looking for.
188 00:21:16.770 ⇒ 00:21:21.840 Awaish Kumar: Then I understand… I have got to understand your… background.
189 00:21:22.020 ⇒ 00:21:28.710 Awaish Kumar: My next few questions are more, like, technical, like, just regarding… Okay.
190 00:21:29.470 ⇒ 00:21:38.600 Awaish Kumar: SQL, Python, And, yeah, and maybe some of the tools… We might have ghost, right?
191 00:21:39.150 ⇒ 00:21:44.969 Awaish Kumar: Okay, sure. So, how would you rate yourself in SQL and Python?
192 00:21:47.130 ⇒ 00:21:54.490 Ujval Kamath: So, I think in terms of the data science, like, if you look at, sort of, data science-focused SQL and Python, I would rate myself pretty well.
193 00:21:54.610 ⇒ 00:21:58.070 Ujval Kamath: Compared to, like, a software engineer, I’m maybe okay.
194 00:21:58.070 ⇒ 00:21:58.870 Awaish Kumar: Like, out of terror.
195 00:21:58.870 ⇒ 00:21:59.480 Ujval Kamath: use a lot.
196 00:21:59.480 ⇒ 00:22:00.370 Awaish Kumar: Read it.
197 00:22:01.390 ⇒ 00:22:03.470 Ujval Kamath: Maybe, like, a 7 on both.
198 00:22:04.060 ⇒ 00:22:09.560 Awaish Kumar: Okay, so, like… what are… Some…
199 00:22:09.840 ⇒ 00:22:16.020 Awaish Kumar: What is the, like, concept of… using analytical functions in SQL.
200 00:22:18.120 ⇒ 00:22:21.819 Ujval Kamath: So, when you say analytical functions, you mean, like.
201 00:22:22.100 ⇒ 00:22:28.350 Ujval Kamath: aggregates, window functions, could you… like, I don’t know the term analytical functions.
202 00:22:28.450 ⇒ 00:22:37.049 Awaish Kumar: Okay, so aggregates are aggregates, right? And analytical functions are like this, where we are going to lose this partition by
203 00:22:37.460 ⇒ 00:22:38.320 Awaish Kumar: Right?
204 00:22:38.550 ⇒ 00:22:42.640 Ujval Kamath: Oh, you mean, like, if you… like a window function, basically?
205 00:22:43.160 ⇒ 00:22:48.189 Awaish Kumar: Yeah, like, all… all the window functions are basically part of analytical functions.
206 00:22:48.630 ⇒ 00:22:55.900 Ujval Kamath: Okay. So, I mean, I haven’t used one recently, but I do know, like, typically, you would,
207 00:22:57.480 ⇒ 00:22:58.530 Ujval Kamath: if…
208 00:22:58.640 ⇒ 00:23:06.690 Ujval Kamath: you know, you can, like, do a partition by, you look at bracket, put in the order, and things like that. I haven’t used a partition function in actually quite a while, but…
209 00:23:06.690 ⇒ 00:23:15.470 Awaish Kumar: And, okay, then what is… okay, what do you think the… concept of… cities.
210 00:23:15.690 ⇒ 00:23:16.940 Awaish Kumar: What is that?
211 00:23:19.070 ⇒ 00:23:20.809 Ujval Kamath: Custom table expressions?
212 00:23:21.500 ⇒ 00:23:23.120 Awaish Kumar: common table expressions.
213 00:23:23.120 ⇒ 00:23:27.849 Ujval Kamath: Commentation, yeah. So this is where I… I’m telling you very honestly, like, I’m not…
214 00:23:27.990 ⇒ 00:23:39.589 Ujval Kamath: I have done a lot of work downstream of some of this stuff, like CTEs and all this, is typically more of a part of data engineering. So that’s why I said, like, I think in terms of SQL for data analysis, so if I have to do
215 00:23:39.660 ⇒ 00:23:49.810 Ujval Kamath: a sort of subquery and things like that. I’m pretty good if I have to write my own SQL, but I have never written a CTE in my life, to be honest.
216 00:23:50.280 ⇒ 00:23:55.549 Awaish Kumar: Okay, that… I think that’s just an improved subcarriery. So, if you have used subcarri.
217 00:23:55.980 ⇒ 00:24:00.600 Awaish Kumar: So the way to optimize that is using CTEs.
218 00:24:01.480 ⇒ 00:24:02.100 Ujval Kamath: Okay.
219 00:24:02.450 ⇒ 00:24:10.120 Ujval Kamath: Okay. Yeah, I, like I said, I have definitely used subqueries within a query, but I have never used the CTE. I’ve never written one.
220 00:24:11.540 ⇒ 00:24:16.279 Awaish Kumar: Okay, have you worked with databases?
221 00:24:17.280 ⇒ 00:24:17.870 Ujval Kamath: Yes.
222 00:24:18.290 ⇒ 00:24:19.090 Ujval Kamath: Many.
223 00:24:20.340 ⇒ 00:24:22.600 Awaish Kumar: Teradata, SQL…
224 00:24:22.640 ⇒ 00:24:26.780 Ujval Kamath: Other, like, Postgres, the…
225 00:24:27.320 ⇒ 00:24:34.729 Awaish Kumar: Okay, I’ll, what are some optimization techniques you would do in Postgres.
226 00:24:36.520 ⇒ 00:24:37.230 Awaish Kumar: They’re funny.
227 00:24:37.230 ⇒ 00:24:40.989 Ujval Kamath: In terms of the… for the SQL query, or for the on-the-table?
228 00:24:40.990 ⇒ 00:24:43.970 Awaish Kumar: So, Postgres, we have a Postgres database.
229 00:24:44.160 ⇒ 00:24:50.130 Awaish Kumar: There are some tables, and then you are writing some queries to access the data, right?
230 00:24:50.360 ⇒ 00:24:53.639 Awaish Kumar: But that is slow. You are getting your response when…
231 00:24:54.260 ⇒ 00:25:11.179 Awaish Kumar: like, it is not responsive, you are… it is taking, like, a few minutes to come up with some answer. So, how… what are some optimization techniques you would do? It does not matter if… if you are optimizing the table, and also doesn’t matter if you’re optimizing the query itself.
232 00:25:12.260 ⇒ 00:25:20.160 Ujval Kamath: Sure. So, I mean, from a table search perspective, I think if I looked at the… if I looked at the query, and I saw that, you know, there was, like, a where clause or some sort of search.
233 00:25:20.160 ⇒ 00:25:34.459 Ujval Kamath: The first thing I would do is, just make sure that there are indexes on, you know, on those particular columns, so that that would optimize it. I would try to make sure they’re not doing something like a full table join, or something that requires, like, a full scan of the table.
234 00:25:34.520 ⇒ 00:25:36.810 Ujval Kamath: That would be in the query.
235 00:25:37.070 ⇒ 00:25:47.230 Ujval Kamath: And, you know, if they are doing, like, sometimes people will do something like a full join, when they could actually use something like… like, I’ve seen this myself, where, like, somebody is doing, like, this big join.
236 00:25:47.330 ⇒ 00:25:57.640 Ujval Kamath: But you can actually write a subquery, which, like, maybe caches part of it, like, the join columns, so that way, it’s… instead of looking, like, at the whole table, you get, like, a summarized
237 00:25:57.790 ⇒ 00:26:04.460 Ujval Kamath: view of just, like, the matching columns. Another way, I think, would be…
238 00:26:04.460 ⇒ 00:26:05.060 Awaish Kumar: That’s subcater.
239 00:26:05.060 ⇒ 00:26:05.900 Ujval Kamath: Let’s just…
240 00:26:07.590 ⇒ 00:26:15.029 Ujval Kamath: No, so, okay, so let’s just say that you have a table with a lot of duplicates. So, so let’s just say that.
241 00:26:15.370 ⇒ 00:26:21.289 Awaish Kumar: Yeah, that is removing duplicates in a subcarrierry, but… Yeah, yeah.
242 00:26:22.350 ⇒ 00:26:24.200 Ujval Kamath: No, so what I’m saying is…
243 00:26:24.340 ⇒ 00:26:32.139 Ujval Kamath: So, I didn’t mean you’re caching a subquery, I mean, you’re kind of using it like a temporary cache. So, like, you write a subquery which removes the duplicates, and then you join on that.
244 00:26:32.140 ⇒ 00:26:32.650 Awaish Kumar: Yeah.
245 00:26:32.650 ⇒ 00:26:33.580 Ujval Kamath: So you’re…
246 00:26:33.580 ⇒ 00:26:35.719 Awaish Kumar: That’s just filtering the data, right? So…
247 00:26:36.270 ⇒ 00:26:39.110 Awaish Kumar: Before you’d make out of it.
248 00:26:39.110 ⇒ 00:26:49.180 Ujval Kamath: before you do a join. So the join is simpler, you know, it’s not doing, like, a double join on a whole table. Another thing is maybe, like, take some practical steps. So, for example, if…
249 00:26:49.470 ⇒ 00:26:52.930 Ujval Kamath: People are like, oh, we query this data every morning.
250 00:26:53.340 ⇒ 00:26:57.299 Ujval Kamath: And, we only use it at, like, 10 a.m. once.
251 00:26:57.450 ⇒ 00:27:00.019 Ujval Kamath: Maybe you can create, like, a materialized view.
252 00:27:00.350 ⇒ 00:27:11.159 Ujval Kamath: From, like, a scheduled materialized view, and then… so that way, when they open in the morning, even if the query takes, like, a little bit of time to run, they won’t see it, you know, they just access the…
253 00:27:11.370 ⇒ 00:27:13.599 Ujval Kamath: The actual, like, finalized view.
254 00:27:13.990 ⇒ 00:27:17.799 Awaish Kumar: Okay. I mean, that’s… Okay, is there anything else you can do?
255 00:27:20.640 ⇒ 00:27:24.700 Ujval Kamath: I can’t think of anything off the top of my head, no.
256 00:27:25.010 ⇒ 00:27:27.150 Awaish Kumar: Okay. Can you do partitioning?
257 00:27:30.860 ⇒ 00:27:37.140 Ujval Kamath: You mean parti… you mean actually, like… okay, so you don’t mean, like… you mean, like, actually, like, partitioning the data on the…
258 00:27:37.370 ⇒ 00:27:41.610 Ujval Kamath: On the… on the database, so, like, it’s, like, partitioned by date, so it, like, runs…
259 00:27:41.610 ⇒ 00:27:50.729 Awaish Kumar: It’s a table, right? It’s a table, right? For the table, you have to mention for things, like, you can… you are going to verify if column has the indexes, right?
260 00:27:50.750 ⇒ 00:28:07.180 Awaish Kumar: So, similarly, you can also, on the table, you can put partitioning, put electric partitioning technique on some column, which… where you see the filters, right? So, basically, if I’m looking at the date filter, and I have a… normally my queries are… Right.
261 00:28:07.230 ⇒ 00:28:09.589 Awaish Kumar: For the recent months.
262 00:28:09.930 ⇒ 00:28:21.250 Awaish Kumar: your partition… like, partitioning will help you optimize the scan, how much data you are going to scan, right? So, it is also going to solve
263 00:28:21.520 ⇒ 00:28:25.920 Awaish Kumar: Your issue of… Thanks a lot, Curtis.
264 00:28:27.330 ⇒ 00:28:31.320 Ujval Kamath: Yeah, I mean, I was aware of Sort of using partitioning, so…
265 00:28:31.690 ⇒ 00:28:33.710 Ujval Kamath: You know, let’s just say that you have
266 00:28:35.250 ⇒ 00:28:42.199 Ujval Kamath: you know, like a multi-core system. You can partition just to say that, okay, we can partition by month, but I mean, I understand what you’re saying.
267 00:28:43.680 ⇒ 00:28:44.270 Awaish Kumar: Yeah.
268 00:28:44.550 ⇒ 00:28:45.740 Awaish Kumar: Yeah, that…
269 00:28:45.910 ⇒ 00:28:52.360 Awaish Kumar: That’s a different concept, right? We’re not in Spark, right? It’s where you want to be partitioning to…
270 00:28:53.380 ⇒ 00:28:58.080 Awaish Kumar: to, like, send the load to multiple systems, database and tables.
271 00:28:58.700 ⇒ 00:28:59.270 Ujval Kamath: Right.
272 00:29:01.380 ⇒ 00:29:02.260 Awaish Kumar: Wow.
273 00:29:03.130 ⇒ 00:29:15.740 Awaish Kumar: And, yeah, so that’s what you can do. And in terms of indexing, did you know, like, how that works? What is, clustered index and non-clustered index?
274 00:29:17.210 ⇒ 00:29:23.310 Ujval Kamath: Yeah, like I said, I’m not that strong in data warehousings. Typically, what I would do is, if I had to set an index on something, I would go in.
275 00:29:23.460 ⇒ 00:29:30.279 Ujval Kamath: And just update the table to set an index, and I… I’m really not that familiar with, sort of, the data warehouse side of things.
276 00:29:31.210 ⇒ 00:29:38.070 Awaish Kumar: Yeah, but okay, that, that is, like, common concepts in databases and data warehouses.
277 00:29:38.710 ⇒ 00:29:48.870 Awaish Kumar: So, apart from that, what… Okay, if I, like… If you summarize yourself.
278 00:29:49.310 ⇒ 00:29:56.070 Awaish Kumar: that I’m expert in… Like, one to three things, like, what could those be?
279 00:29:57.480 ⇒ 00:30:04.100 Ujval Kamath: I mean, I think I would consider myself an expert on taking, like, a data science project from conception to production.
280 00:30:05.180 ⇒ 00:30:05.660 Awaish Kumar: There are some…
281 00:30:05.660 ⇒ 00:30:06.310 Ujval Kamath: there’s…
282 00:30:06.310 ⇒ 00:30:09.420 Awaish Kumar: That’s a really broad thing, a data science project…
283 00:30:09.420 ⇒ 00:30:09.990 Ujval Kamath: Yep.
284 00:30:09.990 ⇒ 00:30:22.530 Awaish Kumar: includes a lot of things, right? The databases optimization is also part of some of data science projects, because you need to store the data somewhere, you will be running queries, and then
285 00:30:22.530 ⇒ 00:30:32.819 Awaish Kumar: maybe you have large amounts of data, and it’s not working, so you might need help of data engineers. So that’s why this… this term is a lot bigger than
286 00:30:32.920 ⇒ 00:30:36.069 Awaish Kumar: Like, it includes a lot of things, so I want to just…
287 00:30:36.280 ⇒ 00:30:44.409 Awaish Kumar: keep yourself, like… I can build a… I’m very good at, like, for example, building a machine learning model, or I’m good at setting up the
288 00:30:44.750 ⇒ 00:30:49.510 Awaish Kumar: for ML models, or… Like, which are, like, kind of narrow?
289 00:30:49.860 ⇒ 00:30:50.620 Awaish Kumar: So…
290 00:30:50.900 ⇒ 00:31:02.079 Ujval Kamath: Sure. So, I mean, I… I think I could set up the infrastructure for ML models for certain parts of it. I, like, I’m not… I don’t have much experience with setting up CICD pipelines.
291 00:31:02.150 ⇒ 00:31:17.750 Ujval Kamath: I’ve certainly used them, but I don’t have much experience setting it up. I think where I would typically set up is if you had some sort of infrastructure for, you know, let’s say you have 100 model, you say… let’s say you’re in a situation where you have, like, 100 models, and they’re all, like, sort of running independently, and you wanted to say.
292 00:31:17.810 ⇒ 00:31:25.430 Ujval Kamath: Okay, build up something that can sort of run these in a more organized way, and set up metrics on it and stuff, like, certainly something I could do.
293 00:31:25.690 ⇒ 00:31:31.479 Ujval Kamath: I think if I had to, like, build them, like, do the data engineering
294 00:31:31.970 ⇒ 00:31:35.410 Ujval Kamath: after the data warehouse, I could certainly do that.
295 00:31:35.700 ⇒ 00:31:42.160 Ujval Kamath: I mean, I’ve done that a lot. If I have to do the ML part, I can do that as a lot. I think one area where
296 00:31:42.600 ⇒ 00:31:48.219 Ujval Kamath: like, let’s just say that I was in your scenario where you said, oh, you… I was writing, like, a lot of
297 00:31:48.930 ⇒ 00:31:52.540 Ujval Kamath: Predictions to a table, and…
298 00:31:52.900 ⇒ 00:32:07.929 Ujval Kamath: typically, like, someone’s doing ad hoc queries the way that you would describe. I would typically, like, go to a data engineer and say, like, how can, you know, I’ve got these issues, I need some help to optimize this table, because that’s not an area where typically I would do it, so…
299 00:32:08.280 ⇒ 00:32:13.979 Awaish Kumar: So, like, the part of ML project is also the feature engineering, right?
300 00:32:14.310 ⇒ 00:32:14.940 Ujval Kamath: Right.
301 00:32:15.150 ⇒ 00:32:24.559 Awaish Kumar: And that… you mentioned that you do that, right? You do the feature engineering and everything. And for feature engineering, you might have to access a lot of data.
302 00:32:25.360 ⇒ 00:32:25.890 Ujval Kamath: Okay.
303 00:32:25.890 ⇒ 00:32:33.950 Awaish Kumar: That could be… there could be, like, maybe, there is some data which is changing,
304 00:32:34.330 ⇒ 00:32:39.100 Awaish Kumar: Every day, and you have to create features. Maybe UML.
305 00:32:39.370 ⇒ 00:32:43.960 Awaish Kumar: Your… the model is dependent on outcome of some other model.
306 00:32:44.070 ⇒ 00:32:56.339 Awaish Kumar: Right? Okay. And that is basically rewriting all their predictions every day. So, for that reason, you might have to rerun all your feature engineering every day.
307 00:32:56.340 ⇒ 00:32:56.940 Ujval Kamath: Okay.
308 00:32:56.940 ⇒ 00:33:13.860 Awaish Kumar: Right? And for that, it’s possible that maybe SQL queries your writing to access your data, create some features, and you’re not able to do that, because the table you are reading from, it is… maybe it is not correctly
309 00:33:14.330 ⇒ 00:33:16.710 Awaish Kumar: Partition, or it does not have good.
310 00:33:16.710 ⇒ 00:33:17.170 Ujval Kamath: Okay.
311 00:33:17.170 ⇒ 00:33:19.019 Awaish Kumar: Correct, indexes, right?
312 00:33:19.490 ⇒ 00:33:23.470 Awaish Kumar: Maybe your curry is really good, like, there’s no problem in the curry.
313 00:33:23.760 ⇒ 00:33:27.449 Awaish Kumar: Maybe problem lies in that table, right? The structure of the table.
314 00:33:27.570 ⇒ 00:33:28.400 Awaish Kumar: So…
315 00:33:28.830 ⇒ 00:33:40.350 Awaish Kumar: that’s, like, you would spot, like, although you would say, like, I’m… I have an optimized security, but it still doesn’t work, right, for me. So how… what would you do? Either if you have
316 00:33:40.350 ⇒ 00:33:51.209 Awaish Kumar: access to that system, you would just go there and optimize that table so that you can improve your… and maybe you are the one who built both models, so you don’t have anywhere…
317 00:33:51.220 ⇒ 00:33:53.290 Awaish Kumar: To tell, right, optimize your table.
318 00:33:53.290 ⇒ 00:33:53.960 Ujval Kamath: Right.
319 00:33:54.550 ⇒ 00:34:08.649 Ujval Kamath: Yeah, I mean, I still think, like, I… it’s not that I wouldn’t optimize it, but I think this is the case where, one is, if it’s not my table, I would never touch it, without talking to, like, whoever table it is, because I don’t know, I will do something, and then somebody else’s query is gonna break.
320 00:34:09.120 ⇒ 00:34:14.060 Ujval Kamath: But I think that’s the case where I think if I was in that situation, I would probably…
321 00:34:14.960 ⇒ 00:34:24.550 Ujval Kamath: do my own research, or reach out to a data engineer and say that, or somebody I know, you know, I’ve worked with, and I say, like, you’re better at this than me, this is what I’m trying to do, can you give me some ideas, and then I would.
322 00:34:24.550 ⇒ 00:34:34.989 Awaish Kumar: Well, yeah, that’s why I came up with this thing that, okay, like, everybody is expert in some areas, right? And that’s… Right.
323 00:34:35.449 ⇒ 00:34:40.030 Awaish Kumar: Understand, right? Okay, maybe… Yeah. Yeah, that’s all.
324 00:34:40.960 ⇒ 00:34:46.619 Ujval Kamath: Yeah, so that’s what I’m saying. I don’t think… I mean, I’m more than honest to say that. I think…
325 00:34:47.889 ⇒ 00:34:56.399 Ujval Kamath: like, optimizing tables, doing those kind of things. I’m not… definitely not an expert on that. I know a little bit, but I think even if… even in, like.
326 00:34:56.690 ⇒ 00:35:16.009 Ujval Kamath: setting an index on a table, unless it was my table, and I was very confident of, like, who’s using it after me, I would probably still reach out to somebody and say that, hey, I want to do these optimizations. Do you think this is even a good idea? Should I do it? Do you have any other ideas of how I can… Definitely that side of it, I’m not… I’m not…
327 00:35:16.010 ⇒ 00:35:16.690 Awaish Kumar: Okay.
328 00:35:16.690 ⇒ 00:35:18.440 Ujval Kamath: I would not consider my expert at all.
329 00:35:19.000 ⇒ 00:35:24.200 Awaish Kumar: As a data scientist, like, can you give me an example of one project
330 00:35:24.650 ⇒ 00:35:27.909 Awaish Kumar: Which you done, from beginning to the…
331 00:35:28.080 ⇒ 00:35:30.299 Awaish Kumar: Like, you did, like, end-to-end thing?
332 00:35:30.480 ⇒ 00:35:36.350 Awaish Kumar: Right. So, what… actually, what steps you followed, and what was your final deliverable?
333 00:35:38.320 ⇒ 00:35:46.650 Ujval Kamath: Sure. So, I worked on this, I worked on this… I have worked on a lot of what’s called predictive maintenance use cases.
334 00:35:46.850 ⇒ 00:35:52.219 Ujval Kamath: So, I worked what… one is I worked for this medic… I worked for a medical device company.
335 00:35:52.500 ⇒ 00:36:04.339 Ujval Kamath: And typically, these medical devices are really complicated, so they’re complex mechanically and, like, electronically. And typically when they look for failure, what they do is.
336 00:36:05.350 ⇒ 00:36:24.210 Ujval Kamath: They… what they basically do is they… when they fail, they… it’s a big issue. Like, you can imagine, like, if you go to a hospital and the machine is broken, you don’t want that to be the case, right? When you go as a patient, or if you’re a doctor, you want the machine to work. So typically, when these… there’s different kind of ways that these machines fail.
337 00:36:24.400 ⇒ 00:36:35.649 Ujval Kamath: And the company I was working for basically wanted to sort of detect, when there was machines with the… issues with the machine before it sort of manifested in the machine, like, not working.
338 00:36:35.790 ⇒ 00:36:43.329 Ujval Kamath: So, you can think about, like, an example of this is, let’s just say that there is a fluid system that, like, puts fluids in the machine.
339 00:36:43.530 ⇒ 00:36:48.060 Ujval Kamath: it starts leaking. It’s just one drop per day, right? There’s, like, a small leak.
340 00:36:49.660 ⇒ 00:36:59.330 Ujval Kamath: If you can capture that when it’s one drop per day, as opposed to capturing it when, like, the whole pump is leaking and all the fluid is coming out of the bottom of the machine, that’s really… that can be really important.
341 00:36:59.470 ⇒ 00:37:04.839 Ujval Kamath: So, typically, this company had a lot of historical, like, service data.
342 00:37:05.000 ⇒ 00:37:07.349 Ujval Kamath: So, they… they had, like.
343 00:37:07.660 ⇒ 00:37:16.400 Ujval Kamath: a list of when the machine had broken in the past. They also had a lot of live streaming data from the machine, so log data, sensor data, and things like that. So…
344 00:37:16.670 ⇒ 00:37:29.890 Ujval Kamath: what I did was I worked with somebody who was, like, a mechanical engineer, and they could kind of tell me, like, oh, this is how the machine works, this is how the fluid system works, and what that allowed me to do is understand, like, what is the right data that I need to, like.
345 00:37:30.110 ⇒ 00:37:32.639 Ujval Kamath: This machine has, like, lots of data coming in.
346 00:37:32.770 ⇒ 00:37:46.229 Ujval Kamath: from, like, every part of it. Okay, what is the right systems I need to focus on? What are the right log messages that sort of filter down? What are the right sensor data? And then ultimately, like, what I did was, like, build a data pipeline.
347 00:37:46.380 ⇒ 00:37:51.450 Ujval Kamath: That basically looked at, okay, let’s just give you an example. Let’s just say that the
348 00:37:51.880 ⇒ 00:37:53.940 Ujval Kamath: System fail… let’s just say that the…
349 00:37:54.100 ⇒ 00:38:05.380 Ujval Kamath: a leak happened, was detected, and, like, somebody… the issue occurred on January 31st, right? Let me… actually, let me give you a better date. Let’s say May 31st, right? So…
350 00:38:05.930 ⇒ 00:38:15.620 Ujval Kamath: Ideally, what you want to do is, can you detect it on May 1st before, like, it really has an issue on the machine? So then, take the sensor data, take the log data from
351 00:38:15.890 ⇒ 00:38:18.030 Ujval Kamath: april.
352 00:38:18.900 ⇒ 00:38:35.479 Ujval Kamath: and then essentially build, like, features on that. So a typical feature would be, like, what is the daily average value, what is the… of this fluid pump, what is the sensor… what is, like, you know, do these kind of aggregate… build these sort of feature stores. In terms of what’s the deliverable, is that
353 00:38:35.830 ⇒ 00:38:37.699 Ujval Kamath: So the iteration is, like.
354 00:38:37.700 ⇒ 00:38:41.059 Awaish Kumar: I also want to understand what are the tools being used, what are the…
355 00:38:41.060 ⇒ 00:38:41.490 Ujval Kamath: Sure.
356 00:38:41.490 ⇒ 00:38:43.039 Awaish Kumar: Which forms being utilized?
357 00:38:43.730 ⇒ 00:38:51.639 Ujval Kamath: Sure. So, in terms of the data pipeline, it’s almost all Python with Pandas or PySpark?
358 00:38:51.820 ⇒ 00:39:09.500 Ujval Kamath: So typically, what I did was… the amount of sensor data that we had was quite large. So, you know, it was, like, 5-second resolution. The log data is, like, just as voluminous, so typically all this data was stored in all Delta Lake tables.
359 00:39:09.710 ⇒ 00:39:14.539 Ujval Kamath: So, we query those Delta Lake tables, use PySpark to essentially filter out
360 00:39:14.800 ⇒ 00:39:19.320 Ujval Kamath: and get it to the point where, like, it’s more manageable in Pandas.
361 00:39:19.930 ⇒ 00:39:21.060 Ujval Kamath: And then…
362 00:39:21.270 ⇒ 00:39:28.979 Ujval Kamath: Using, sort of, simple pandas functions, so, you know, they have, like, groupys and, like, time series aggregates aggregated into days.
363 00:39:29.320 ⇒ 00:39:41.240 Ujval Kamath: And then, essentially, that sort of becomes features. And the machine learning is just done using scikit-learn. So the deployed model is essentially a similar… is, like, a very, very similar script.
364 00:39:42.380 ⇒ 00:39:47.270 Awaish Kumar: No, no, but what… What was the outcome, like, of the model? What was the model?
365 00:39:47.270 ⇒ 00:39:57.140 Ujval Kamath: And so the outcome is essentially, like, about 2 to 4 weeks before this event occurs, the model generates, you know, it’s continuously generating predictions.
366 00:39:57.290 ⇒ 00:40:10.140 Ujval Kamath: And then, if the… if, like, it predicts… it’s essentially a classification algorithm, so it predicts, like, zero, like, we don’t detect… we don’t predict some sort of leak issue, or 1, like, it predicts a leak issue. That one is then…
367 00:40:10.480 ⇒ 00:40:17.519 Ujval Kamath: sent to a sub… like, in that case, what happens is that the model, all the… what we call them alarms, they’re essentially
368 00:40:18.100 ⇒ 00:40:18.880 Ujval Kamath: an issue.
369 00:40:20.320 ⇒ 00:40:21.320 Ujval Kamath: Go ahead.
370 00:40:21.320 ⇒ 00:40:23.229 Awaish Kumar: How the model was, deployed.
371 00:40:24.760 ⇒ 00:40:36.720 Ujval Kamath: So the deploy… it was deployed in Databricks, and it ran, like, 6 times a day. So every 4 hours, it would query, like, the last 4 hours of data, and then it would essentially,
372 00:40:37.460 ⇒ 00:40:40.939 Ujval Kamath: It would essentially, generate a prediction.
373 00:40:41.060 ⇒ 00:40:44.929 Ujval Kamath: There was, like, a control process that ran it 4 times a day.
374 00:40:45.060 ⇒ 00:40:46.460 Ujval Kamath: I mean, 6 times a day.
375 00:40:46.620 ⇒ 00:40:49.660 Awaish Kumar: Yeah, but my question is that, for example, as you mentioned.
376 00:40:49.970 ⇒ 00:40:57.450 Awaish Kumar: So you have some Spark scripts, you also have some model deployed somewhere, which is running on a schedule.
377 00:40:58.060 ⇒ 00:40:59.950 Awaish Kumar: How it is running, like…
378 00:41:00.520 ⇒ 00:41:08.609 Awaish Kumar: what is the… what is the orchestration tool? How… who’s scheduling those after every, 4-hour?
379 00:41:10.170 ⇒ 00:41:13.539 Ujval Kamath: So there was only one or… there was one airflow,
380 00:41:13.910 ⇒ 00:41:26.430 Ujval Kamath: DAG, that essentially, essentially what happened was, like, the Databricks job, it was exposed as the Databricks job, so there was one central control job that essentially called each individual model.
381 00:41:26.500 ⇒ 00:41:35.570 Ujval Kamath: Okay. And the DAG script was a… the Airflow DAG would essentially just run every 4 hours. It would call that… that… that sort of Databricks job.
382 00:41:35.880 ⇒ 00:41:41.260 Ujval Kamath: And then the individual models would write the,
383 00:41:42.010 ⇒ 00:41:45.569 Ujval Kamath: Positive predictions into a downstream system.
384 00:41:45.820 ⇒ 00:41:48.469 Ujval Kamath: And that was then used by the service people.
385 00:41:48.870 ⇒ 00:41:59.110 Ujval Kamath: The downstream system was an… was an Azure SQL database. That Azure SQL database, once per day, or… I don’t remember it. It used to then go pushed into, like, a…
386 00:41:59.360 ⇒ 00:42:05.409 Ujval Kamath: a system that got pushed into SAP. And I was not… we were not touching the SAP system at all, so…
387 00:42:08.880 ⇒ 00:42:10.740 Awaish Kumar: Okay, got it.
388 00:42:10.980 ⇒ 00:42:12.970 Awaish Kumar: So, yeah.
389 00:42:13.800 ⇒ 00:42:16.090 Awaish Kumar: I understand the complete flow now.
390 00:42:16.270 ⇒ 00:42:20.710 Awaish Kumar: And so, what were the… what was your… wherever you’re…
391 00:42:21.580 ⇒ 00:42:24.539 Awaish Kumar: contributions? Like, were you a part of
392 00:42:24.670 ⇒ 00:42:31.940 Awaish Kumar: the team which is sending data from those machines to some data lake. Were you part of…
393 00:42:32.480 ⇒ 00:42:39.460 Awaish Kumar: like, the team or yourself, which is sending from data layer, reading from data layer, creating PySpark.
394 00:42:39.590 ⇒ 00:42:45.660 Awaish Kumar: Scripts to basically… Do some filtering or whatever, and then…
395 00:42:46.110 ⇒ 00:42:50.550 Awaish Kumar: Were you at the point where you are doing Panda’s exploration?
396 00:42:50.950 ⇒ 00:42:54.519 Awaish Kumar: Or, at the deployment, like… like, there’s a…
397 00:42:56.300 ⇒ 00:42:59.950 Awaish Kumar: like, there is a lot of steps in this pipeline, so I want to understand.
398 00:42:59.950 ⇒ 00:43:00.440 Ujval Kamath: Sure.
399 00:43:00.440 ⇒ 00:43:04.669 Awaish Kumar: At what… what steps that… which steps are the ones which
400 00:43:04.790 ⇒ 00:43:07.829 Awaish Kumar: You were doing, and what are the other team members were doing?
401 00:43:09.150 ⇒ 00:43:20.010 Ujval Kamath: Sure. So, I didn’t collect data from the machines. There was, sort of a cloud team and a data engineering team that put it in Delta Lake tables. I…
402 00:43:20.360 ⇒ 00:43:32.900 Ujval Kamath: I was not the Airflow person either. I was an admin on the Airflow table, on the Airflow system, so, like, if it failed, or if there was an issue, I would get an alert. I would go and, like, I could restart it, but typically,
403 00:43:33.130 ⇒ 00:43:43.530 Ujval Kamath: there was an Airflow team who was in charge of those DAGs. I could certainly build them myself, but I never needed to, because it was just, like, one DAG that called that system. But after the data warehouse.
404 00:43:43.720 ⇒ 00:43:44.700 Ujval Kamath: to…
405 00:43:45.420 ⇒ 00:43:55.020 Ujval Kamath: deploying, like, that whole thing, which was, like, writing the PySpark code, writing the Python code, building the model, putting it into the production system, that was all me.
406 00:43:56.180 ⇒ 00:43:57.699 Awaish Kumar: Okay. But that was just me.
407 00:43:57.930 ⇒ 00:44:03.680 Ujval Kamath: like I said, it was a pretty small team. We didn’t have, like, a dedicated, like, so many dedicated people to do all these things.
408 00:44:04.440 ⇒ 00:44:10.940 Awaish Kumar: So you have a… you were writing the Spark jobs, and then you were writing the… Huh.
409 00:44:11.230 ⇒ 00:44:15.120 Awaish Kumar: Bondo’s, and then you were also…
410 00:44:15.240 ⇒ 00:44:23.550 Awaish Kumar: Kind of working on the model that basically fine-tuning the model and all of that, and then finally deploying it, right?
411 00:44:23.550 ⇒ 00:44:24.220 Ujval Kamath: Yes.
412 00:44:24.810 ⇒ 00:44:28.099 Ujval Kamath: So, I mean, Pi… yeah, it was all PySpark, yeah, but yeah.
413 00:44:30.250 ⇒ 00:44:32.309 Awaish Kumar: You mentioned Panda also, right?
414 00:44:33.090 ⇒ 00:44:41.849 Ujval Kamath: Yeah, Pi… so typically, like I said, PySpark to start with, once the data sort of shrunk to a reasonable size, it was Pandas.
415 00:44:42.050 ⇒ 00:44:45.360 Ujval Kamath: once Panda… you can use Pandas through the feature engineering.
416 00:44:45.360 ⇒ 00:44:45.720 Awaish Kumar: Yeah.
417 00:44:45.720 ⇒ 00:44:49.570 Ujval Kamath: Then, it was… The model, you know, the model part.
418 00:44:50.150 ⇒ 00:44:53.569 Ujval Kamath: And then… the deployment.
419 00:44:53.760 ⇒ 00:44:54.360 Awaish Kumar: Okay.
420 00:44:54.710 ⇒ 00:44:59.469 Awaish Kumar: So I just want to understand… that,
421 00:45:01.020 ⇒ 00:45:06.499 Awaish Kumar: Yeah, and that’s… like, that is still all backend, so…
422 00:45:07.230 ⇒ 00:45:14.470 Awaish Kumar: Although, like, you have worked… you have did a lot of work, right, right now, but this is now on the back end, so how…
423 00:45:14.770 ⇒ 00:45:19.740 Awaish Kumar: Where you then finally Showcasing your work to the client.
424 00:45:21.290 ⇒ 00:45:30.330 Ujval Kamath: So, typically with these cases, like, we did a lot of upfront work. So, for example, if we identified one particular part that was failing.
425 00:45:30.530 ⇒ 00:45:35.019 Ujval Kamath: typically, we… I would do a lot of upfront work with
426 00:45:35.290 ⇒ 00:45:44.520 Ujval Kamath: some of the business owners to understand, like, what is the cost-benefit of predicting that this thing is going to fail. So they might say something like, okay, if…
427 00:45:45.050 ⇒ 00:45:57.679 Ujval Kamath: We, let’s say the machine breaks and we have to send somebody in to fix it. They sort of understood from, like, a time and materials, in terms of the machine being down, they understood what the cost of that was.
428 00:45:57.770 ⇒ 00:46:05.590 Ujval Kamath: Like, so, I would work with them, and they would say, okay, what is the cost of that? Then, it would be like, okay, we go back, but let’s say we do it a month from before.
429 00:46:05.640 ⇒ 00:46:18.760 Ujval Kamath: like, say we’re predicting in advance. They understood the cost of doing it that way as well. So what is the cost savings of us doing it that way as opposed to doing it, you know, like, going in at the last minute after the machine is broken? This is what… so then… so then I would…
430 00:46:18.860 ⇒ 00:46:32.859 Ujval Kamath: that’s sort of how I presented the results at the end, was we could say that, okay, over the course of 6 months, we had this many sort of… if you look at the historical data, over the course of 6 months, we had this many issues that occurred.
431 00:46:33.300 ⇒ 00:46:37.349 Ujval Kamath: We were able to predict, like, 75% of them.
432 00:46:37.600 ⇒ 00:46:57.409 Ujval Kamath: So that easily gives us an easy mapping of what is the cost-benefit of doing this. So then when you go to, like, sort of the stakeholders, you say, okay, look, looking at our historical data, we understand that if this model had been in place, this is what we would have approximated for our cost savings, and this is sort of how it would go from a… if we move forward.
433 00:46:58.900 ⇒ 00:47:03.679 Awaish Kumar: Yeah, so, like, I get it, like, you…
434 00:47:03.970 ⇒ 00:47:09.439 Awaish Kumar: came up with some scenarios where… which… where you can actually make an impact in terms of
435 00:47:09.730 ⇒ 00:47:15.350 Awaish Kumar: For the business, basically. But my question was more like, how…
436 00:47:16.030 ⇒ 00:47:23.620 Awaish Kumar: like, for the executives, how were you showcasing that? Like, maybe that was a presentation, or maybe that was a.
437 00:47:23.620 ⇒ 00:47:25.889 Ujval Kamath: Oh, yeah, it’s typically a PowerPoint.
438 00:47:26.070 ⇒ 00:47:26.690 Awaish Kumar: Yeah.
439 00:47:26.840 ⇒ 00:47:30.660 Ujval Kamath: So, I mean, executives don’t… Are you finding time building, or…
440 00:47:32.420 ⇒ 00:47:37.290 Ujval Kamath: Yeah, so typically, it’s just like a PowerPoint… PowerPoint chart with some financials in it?
441 00:47:37.640 ⇒ 00:47:52.370 Ujval Kamath: I mean, like, a PowerPoint with some financials in it, so we would give a little bit of history on the use case, like, maybe one slide, talk about, you know, some basic things, like, okay, we can predict, you know, depending on the issue, like, sometimes we could predict something two weeks in advance, something could be four weeks in advance.
442 00:47:53.850 ⇒ 00:48:13.589 Ujval Kamath: just some background on the use case, and then we could say that, you know, once this goes in the production, we could expect this much cost savings based on… and typically, like, to be, like, honest, like, executives, they’re… my experience with them is, like, you just have a few slides, what is the, sort of, the cost to the benefit to them, and then beyond that, they don’t care. I mean…
443 00:48:14.110 ⇒ 00:48:20.620 Awaish Kumar: And, okay, what are the… like, have you worked with dbt, Snowflake?
444 00:48:23.220 ⇒ 00:48:27.010 Ujval Kamath: I’ve worked with Snowflake, typically just to query data out of it.
445 00:48:27.110 ⇒ 00:48:28.210 Awaish Kumar: Okay.
446 00:48:28.340 ⇒ 00:48:30.600 Ujval Kamath: I have not worked with DPT.
447 00:48:31.670 ⇒ 00:48:42.510 Awaish Kumar: Okay, have you… Yeah, so… How would you rate yourself?
448 00:48:43.020 ⇒ 00:48:53.330 Awaish Kumar: In terms of… Like, working in a client-facing setup, right? So, are you comfortable that…
449 00:48:53.750 ⇒ 00:49:04.340 Awaish Kumar: With presenting whatever your work is, making presentations, like, exec-level presentations, and going in front of execs to, like, showcase your work.
450 00:49:04.670 ⇒ 00:49:09.450 Awaish Kumar: How comfortable you are with that, or… Yeah.
451 00:49:10.590 ⇒ 00:49:21.799 Ujval Kamath: I mean, I would rate myself an 8 or a 9. I spend a lot of years in customer-facing roles, and I’ve done a lot of presentations, to a lot of, like, technical people, non-technical people, executives, so I’m…
452 00:49:22.120 ⇒ 00:49:24.149 Ujval Kamath: I’m comfortable with that, that’s not…
453 00:49:24.150 ⇒ 00:49:26.340 Awaish Kumar: I think that’s it. I…
454 00:49:26.580 ⇒ 00:49:34.870 Awaish Kumar: Yeah, and my last question, like, what are you… what are your expectations, like, in terms of your career growth, like, what are you looking for?
455 00:49:36.480 ⇒ 00:49:47.369 Ujval Kamath: So, I mean, I like customer-facing work. I mean, I like whether it’s stake, whether you want to call them customer… internal customers or external customers. I do like that kind of work. I…
456 00:49:47.630 ⇒ 00:49:51.570 Ujval Kamath: In terms of career growth, I just maybe look for more impact.
457 00:49:52.000 ⇒ 00:49:57.990 Ujval Kamath: Oh, that’s… I… maybe, like, a broader scope on my projects, that’s sort of the main thing.
458 00:49:59.170 ⇒ 00:49:59.950 Awaish Kumar: I’m good.
459 00:50:01.490 ⇒ 00:50:06.250 Awaish Kumar: I mean, like, where do you see yourself in next… 5 years.
460 00:50:07.560 ⇒ 00:50:27.140 Ujval Kamath: I mean, my honest answer is, like, I really don’t know. I… it’s like a question I don’t really have a very good… I don’t have a clear vision. I never really have a clear vision. I like working on data science things, and my… hopefully, for the next 5 years, I can… I can continue to do that, and AI doesn’t completely, like, upturn the whole data science economy.
461 00:50:27.160 ⇒ 00:50:30.160 Ujval Kamath: That’s… that’s kind of where I… I am with that.
462 00:50:30.710 ⇒ 00:50:31.480 Awaish Kumar: Okay.
463 00:50:31.610 ⇒ 00:50:32.160 Awaish Kumar: Oh.
464 00:50:32.160 ⇒ 00:50:34.990 Ujval Kamath: I, I was a manager…
465 00:50:35.120 ⇒ 00:50:39.179 Ujval Kamath: at one point, like, I… I was a data science manager,
466 00:50:39.500 ⇒ 00:50:45.630 Ujval Kamath: I think sometimes when this question is, like, where am I growing? I definitely will see myself, like, in terms of
467 00:50:46.120 ⇒ 00:50:56.729 Ujval Kamath: being more of, on the individual contributor track, I don’t really see myself… maybe I could manage a small team, but I don’t see myself becoming, like, a career manager or anything.
468 00:50:56.830 ⇒ 00:50:58.669 Awaish Kumar: I do see myself just…
469 00:50:58.750 ⇒ 00:51:01.290 Ujval Kamath: Climbing this individual contributor track, whatever that means.
470 00:51:01.290 ⇒ 00:51:08.090 Awaish Kumar: And, Like, given that, like, for example, if I ask
471 00:51:08.390 ⇒ 00:51:10.889 Awaish Kumar: Some of your colleagues, or…
472 00:51:11.350 ⇒ 00:51:20.729 Awaish Kumar: past managers, how would you… how would they rate you in terms of your technical abilities, and also
473 00:51:21.030 ⇒ 00:51:26.439 Awaish Kumar: For you as a… as a colleague, how would they rate you read you, yeah?
474 00:51:27.950 ⇒ 00:51:31.209 Ujval Kamath: I would say all my managers would say that I’m strong, technically.
475 00:51:31.580 ⇒ 00:51:34.809 Ujval Kamath: And I would think they’d say that I’m pleasant to work with.
476 00:51:36.170 ⇒ 00:51:36.580 Awaish Kumar: Okay.
477 00:51:36.580 ⇒ 00:51:37.170 Ujval Kamath: So…
478 00:51:37.720 ⇒ 00:51:40.040 Awaish Kumar: Like, how would you rate, like, out of 10?
479 00:51:43.190 ⇒ 00:51:43.720 Awaish Kumar: How would.
480 00:51:43.720 ⇒ 00:51:44.609 Ujval Kamath: I mean, I…
481 00:51:44.610 ⇒ 00:51:47.279 Awaish Kumar: Like, what are… what are you thinking? How would they rate you, right?
482 00:51:48.610 ⇒ 00:51:53.960 Ujval Kamath: I mean, I think that they would probably say that I’m… or 8 or a 9 on both.
483 00:51:54.270 ⇒ 00:51:59.269 Ujval Kamath: Because I… I think I’ve always had, like, good working relationships with my…
484 00:51:59.480 ⇒ 00:52:02.360 Ujval Kamath: I’ve had good working experiences, I’ve always gotten…
485 00:52:02.650 ⇒ 00:52:20.809 Ujval Kamath: I think especially one manager who I… who I got a new manager after my day… after reorg, and I left, like, soon after. We didn’t really work well together. I think all my managers have always been very happy to be having me on the team, and all the colleagues have also always been happy to reach out to me for technical help, or help with other things.
486 00:52:24.020 ⇒ 00:52:34.340 Awaish Kumar: Okay, great. Yeah, that’s all from my side. So, if you have any questions, you can ask, like, if we still have a few minutes, but yeah, that’s all from mine.
487 00:52:35.520 ⇒ 00:52:39.570 Ujval Kamath: Sure. So, I mean, I think the only, question I have was just around,
488 00:52:40.380 ⇒ 00:52:50.860 Ujval Kamath: how do you work… like, how does Brainforge, or how do you typically work with customers? You know, you’re a very distributed team, I don’t know if your data engineers are near you or all around the world, I just typically.
489 00:52:50.860 ⇒ 00:52:51.220 Awaish Kumar: Yes.
490 00:52:51.550 ⇒ 00:52:52.690 Awaish Kumar: Yeah, it’s not gonna work.
491 00:52:53.170 ⇒ 00:52:54.639 Awaish Kumar: We are a remote team.
492 00:52:55.070 ⇒ 00:53:03.380 Awaish Kumar: no offices, like, everybody, like, it’s fully remote, right? Everybody is across… different time zones.
493 00:53:03.600 ⇒ 00:53:07.270 Awaish Kumar: I, myself, is from… I’m from, Pakistan.
494 00:53:07.420 ⇒ 00:53:17.510 Awaish Kumar: We have a few people from India, we have from Philippine, we have people from U.S, Europe, so kind of all over the world.
495 00:53:19.400 ⇒ 00:53:24.859 Awaish Kumar: Typically, I would say we want some type of,
496 00:53:25.020 ⇒ 00:53:27.770 Awaish Kumar: What you say, overlap?
497 00:53:28.430 ⇒ 00:53:32.839 Awaish Kumar: Like, most of our clients are actually in Eastern time zone.
498 00:53:33.510 ⇒ 00:53:36.929 Awaish Kumar: And our CEO is in Central Time.
499 00:53:37.140 ⇒ 00:53:43.859 Awaish Kumar: So, typically, we overlap 4-5 hours with them, like…
500 00:53:44.310 ⇒ 00:53:44.820 Ujval Kamath: Okay.
501 00:53:44.820 ⇒ 00:53:47.599 Awaish Kumar: either with clients or with CO.
502 00:53:47.710 ⇒ 00:53:52.750 Awaish Kumar: But otherwise, it’s, like, it’s…
503 00:53:53.220 ⇒ 00:54:03.549 Awaish Kumar: Yeah, otherwise, like, if you are in the meetings, if you are in all the required meetings with clients, or with your… if you are, like, collaborating with your
504 00:54:03.690 ⇒ 00:54:06.769 Awaish Kumar: Peers, or, like, the things you have to do.
505 00:54:06.890 ⇒ 00:54:09.419 Awaish Kumar: And you’re able to unblock yourself.
506 00:54:09.620 ⇒ 00:54:12.789 Awaish Kumar: And, yeah, then, all good.
507 00:54:14.370 ⇒ 00:54:23.539 Ujval Kamath: Okay. I mean, I guess one question is, you know, typically with my experience with a lot of customer-facing roles, and maybe this is sort of a data science bias, but…
508 00:54:23.890 ⇒ 00:54:24.600 Ujval Kamath: Hop.
509 00:54:25.030 ⇒ 00:54:26.169 Ujval Kamath: Can you slow me?
510 00:54:31.080 ⇒ 00:54:31.820 Ujval Kamath: Hello.