Meeting Title: Brainforge Interview with Harsh Punjabi Date: 2026-04-09 Meeting participants: Awaish Kumar, Harsh Punjabi
WEBVTT
1 00:00:19.630 ⇒ 00:00:21.040 Awaish Kumar: Hi, Hush, how are you?
2 00:00:22.930 ⇒ 00:00:24.669 Harsh Punjabi: Hey Yavish, I’m good, how are you?
3 00:00:25.090 ⇒ 00:00:26.180 Awaish Kumar: I’m good as well.
4 00:00:27.370 ⇒ 00:00:30.239 Awaish Kumar: Where are you located?
5 00:00:30.690 ⇒ 00:00:40.750 Harsh Punjabi: I am, I’m in a smaller city close to Toronto, so I’m… it’s London, London, Ontario. It’s about 2 hours from Toronto, so I just tried Toronto because it’s easier to explain.
6 00:00:41.110 ⇒ 00:00:41.830 Awaish Kumar: Okay.
7 00:00:43.040 ⇒ 00:00:51.229 Awaish Kumar: Okay, so, nice to meet you, and thank you for joining today. So, in this session, we are just going to talk more about,
8 00:00:51.370 ⇒ 00:00:55.529 Awaish Kumar: Your background, and the projects you’ve worked on, and to understand more.
9 00:00:57.170 ⇒ 00:01:03.909 Awaish Kumar: what you have been doing so far, and… and yeah, I’m here to answer if you have any questions regarding Brainforge.
10 00:01:04.250 ⇒ 00:01:07.470 Awaish Kumar: Okay, sounds good. So, yeah, you can introduce yourself.
11 00:01:08.460 ⇒ 00:01:18.920 Harsh Punjabi: Sure, so, I have about 9 years of experience in data so far. I started off as a business analyst in a consulting company called ZS, so that was back when I was in India.
12 00:01:18.920 ⇒ 00:01:29.410 Harsh Punjabi: And, I was supporting a US client, Gilead Sciences, a pharma client. That was very, like, Teradata, SQL, Tableau dashboarding, and, you know, PowerPoint presentation kind of work.
13 00:01:29.410 ⇒ 00:01:34.800 Harsh Punjabi: But there I kind of grew a bit into that role, supported a project for a brand launch.
14 00:01:34.800 ⇒ 00:01:42.049 Harsh Punjabi: For Gilead Sciences. But it was more analytics-oriented. Eventually, I moved… I switched over to American Express.
15 00:01:42.050 ⇒ 00:01:57.790 Harsh Punjabi: There, I started to get more into, you know, the data infrastructure side of things. I was supporting the customer retention analytics team there, but they were built… post-COVID, they started to build out their retention infra… customer retention-related data infra.
16 00:01:57.820 ⇒ 00:02:14.709 Harsh Punjabi: After that, I got an opportunity to work with a startup in Dubai, and that’s where I got a lot of engineering experience, because I was the second member of the data team there, so we were setting up everything from scratch, so a lot of end-to-end work, you know, Fivetran integrations, then developing the data models.
17 00:02:14.710 ⇒ 00:02:19.299 Harsh Punjabi: Creating reporting pipelines, and then finally dashboards, at the consumption layer.
18 00:02:19.310 ⇒ 00:02:34.130 Harsh Punjabi: Once my Canadian immigration papers came through, I moved to Canada around about 3 years ago. So my first role was with a consulting company, like an American small consulting company, but the client was TD Bank, which is one of the top, like, big banks in Canada.
19 00:02:34.130 ⇒ 00:02:52.600 Harsh Punjabi: So there, I worked on two projects. Both of them were expense warehouse related. One was only for their stock trading platform, but then they had a McKinsey project for cutting the budgets across the entire bank. So we prepared their expenses, like, cost allocation expense data warehouse.
20 00:02:52.600 ⇒ 00:03:09.380 Harsh Punjabi: Which basically captured how cost is being allocated, inside the different departments of the bank. Then the next… then I, like, all of these are contract rules, right? So once that contract ended, I, joined Lightspeed, and I’ve been with Lightspeed, for over 18 months now.
21 00:03:09.380 ⇒ 00:03:25.240 Harsh Punjabi: And, here I started off in their product squad, so we were mainly focused on one metric at that point of time, which was GTV, but then the scope of the work expanded. We started to work on payments, and now I’m working as a vertical lead for the hospitality part.
22 00:03:25.240 ⇒ 00:03:39.070 Harsh Punjabi: So, in the hospitality part, most of the use cases that we deal with are product usage, product adoption, a lot of Salesforce integration, ops, finance, we deal with all of these use cases, and I’m building data models for them.
23 00:03:39.070 ⇒ 00:03:51.750 Harsh Punjabi: Throughout this, like, ever since I’ve gotten into analytics engineering-oriented roles, I’ve worked on dbt, and now I’m working on DataForm. Lightspeed switched over to DataForm because of, you know, their comfort with the Google stack.
24 00:03:51.750 ⇒ 00:03:53.690 Harsh Punjabi: But, we use…
25 00:03:53.690 ⇒ 00:04:03.939 Harsh Punjabi: like, I’ve used both dbt and DataForm, and most recently, I’ve been leveraging this thing called business event analysis modeling.
26 00:04:03.940 ⇒ 00:04:17.730 Harsh Punjabi: to kind of build out the data models and design them. And, of course, like, SQL, Python, building out unit tests within CICD, supporting local dashboards at the final consumption layer. So all of those things have been a part of my current role.
27 00:04:17.740 ⇒ 00:04:27.980 Harsh Punjabi: So yeah, I mean, just to summarize, I’ve transitioned from a business analyst to a data analyst to an analytics engineer over the last 9 years, and
28 00:04:28.150 ⇒ 00:04:42.129 Harsh Punjabi: I’m looking for my next role, so brain fart seemed really interesting. I saw… I saw Thom’s post on LinkedIn, and it really resonated with me, and when I saw the role specifically, it turned out, you know, something that seemed to be a good fit for me, so I applied, yeah.
29 00:04:42.610 ⇒ 00:04:43.989 Awaish Kumar: Okay, so how
30 00:04:44.220 ⇒ 00:04:52.879 Awaish Kumar: Like, if you can give me an example of a project, like the end-to-end deliverable of a project in your current role.
31 00:04:53.130 ⇒ 00:04:58.579 Awaish Kumar: So, what… what it could be, and what was the results, and how were… what were your…
32 00:04:58.820 ⇒ 00:05:01.280 Awaish Kumar: Role in delivering that project.
33 00:05:01.280 ⇒ 00:05:01.930 Harsh Punjabi: Okay.
34 00:05:01.930 ⇒ 00:05:26.919 Harsh Punjabi: So, within… so I’ll start with the hospital pod work that I’m doing, because they moved to a hub-and-spoke structure, like, Lightspeed moved to a hub-and-spoke structure. So, within the hospital pod, what we did was, they had some kind of a warehouse built out already where they had a standardized, a conformed, and a gold layer, but it was not consistent, it didn’t have, you know, any of the AE best practices, and there was a lot of metric reputation there as well, because some of the product adoption metrics specifically, which are
35 00:05:26.920 ⇒ 00:05:50.069 Harsh Punjabi: leveraged by both finance and ops. It had conflicts, and there was a bit of inconsistency there. So what we… what my role there was, I, since I was leading the analytics engineering within the pod, I collaborated with the data ops folks. I first mapped out the business process, right? How, like, what are the steps that a user will take when they sign up? What is… what are the linkages between their CRM and their payment side?
36 00:05:50.070 ⇒ 00:05:52.890 Harsh Punjabi: Like, I mapped out the entire business process.
37 00:05:52.890 ⇒ 00:06:05.630 Harsh Punjabi: And, designed the, like, designed the final consumption start schema first, based on business event analysis modeling, where we… it’s a framework, basically. So what… so it basically has the seven W’s.
38 00:06:05.630 ⇒ 00:06:21.640 Harsh Punjabi: And which has one of how much. So, when becomes the time, like, time dimension, where becomes the location dimension, and how much becomes your fact, right? What are you trying to measure? Design the data model, and then implemented that using data form.
39 00:06:21.640 ⇒ 00:06:39.250 Harsh Punjabi: Just like dbt data form also, you can prepare .sqlx files and get your, you know, UDFs integrated, use either source tables or ref tables, like, design the whole thing. And, once the final layer was built.
40 00:06:39.250 ⇒ 00:06:54.509 Harsh Punjabi: we… like, that was fed into the Looker dashboarding. So Looker Dashboard is kind of like a collaborative effort between analytics engineers and the data analysts. So we built out the LookMLs, the LookML layer for the Looker dashboards, and they built out the final visualizations.
41 00:06:54.840 ⇒ 00:07:11.639 Harsh Punjabi: And, within this, we are now trying to integrate some kind of a conversational analytics. So, once our, so, like, I’m working on a solution for this. I’ve done this in Snowflake before. So, Snowflake has Cortex Analyst, where you can build a semantic layer, and then conversational analytics works.
42 00:07:11.640 ⇒ 00:07:28.789 Harsh Punjabi: But now we are exploring VANA AI. So VANA AI would, you know, we’re trying to see if we can train the text-to-SQL logic there. So we’re still solutioning it out, but I’ve built the semantic layer already, and once the model is trained, we can kind of integrate Text2SQL for the stakeholders.
43 00:07:28.790 ⇒ 00:07:32.359 Awaish Kumar: What is the warehouse that you are using right now? Snowflake?
44 00:07:32.360 ⇒ 00:07:47.110 Harsh Punjabi: Right now, right now I’m using BigQuery, so I’ve seen this in Snowflake in my last role, what Cortex Analyst does, but right now, because Lightspeed is fully on the Google stack, so they moved from dbt to DataForm, they’re using BigQuery, so it’s entirely on the Google stack now.
45 00:07:47.110 ⇒ 00:07:51.979 Awaish Kumar: So how… what differences do you see between BigQuery and Snowflake?
46 00:07:52.790 ⇒ 00:07:58.540 Harsh Punjabi: So, I would… okay.
47 00:07:58.720 ⇒ 00:08:15.359 Harsh Punjabi: So, Snowflake is, snowflake has better integrations. If I just put it simply, Snowflake has better integrations, and even though I’ve not built out the backend for Snowflake, so I won’t know… I’m not sure if I can answer that fully, but,
48 00:08:15.510 ⇒ 00:08:22.000 Harsh Punjabi: I mean… Let me rephrase that. So…
49 00:08:25.110 ⇒ 00:08:44.600 Harsh Punjabi: BigQuery is good… BigQuery has a lot of good functions that we can incorporate within the SQL layer, like UnNest and all, which are not present in Snowflake. It’s slightly different. And Snowflake has… Snowflake has, like, good integrations, like Cortex Analyst, and you can build out semantic layers, it has better features in that regard.
50 00:08:44.600 ⇒ 00:08:48.270 Awaish Kumar: have similar features in Snowflake to unnest the JSON.
51 00:08:48.810 ⇒ 00:09:03.430 Harsh Punjabi: Yeah, there are similar features, of course, but, like, I’m just saying that this varies slightly, and, but yeah, like, in my personal opinion, the integrations in BigQuery are not good enough, it’s pretty closed-looped. Snowflake has external integrations as well, yeah.
52 00:09:03.430 ⇒ 00:09:13.469 Awaish Kumar: I can clarify my question. I’m not, like, talking exactly about the function names and the syntaxes, I’m more about, like, what the major difference in terms of
53 00:09:13.600 ⇒ 00:09:29.650 Awaish Kumar: how it processes the query, or how it stores the data, or how the cost optimization works there. If, for example, tomorrow a client comes in, and if we assign you on that client, client has a data warehouse already in the
54 00:09:29.740 ⇒ 00:09:36.399 Awaish Kumar: in the BigQuery, and we need to audit it, optimize the cost. So, like, what are the things that you will,
55 00:09:36.550 ⇒ 00:09:40.889 Awaish Kumar: Like, what are the steps that you would follow or do to audit that?
56 00:09:41.200 ⇒ 00:09:46.059 Awaish Kumar: And then I’m gonna recommend your suggestions for optimization.
57 00:09:46.660 ⇒ 00:09:57.610 Harsh Punjabi: Okay, so, first of all, I would, so I would do, like, a usage audit first, based on their… because you can take the information schema in BigQuery and do the usage audit.
58 00:09:57.610 ⇒ 00:10:08.320 Harsh Punjabi: Based on the usage audit and what type of consumption they have, what type of stakeholders are consuming, what is their cost per query, and, you know, what their overall cloud storage costs are.
59 00:10:08.320 ⇒ 00:10:25.919 Harsh Punjabi: I would, like, compare those against Snowflake and against BigQuery. I would also see if there is any serverless query execution possible within, you know, in either of those two. Azure, like, I’ve seen Azure Synapse as serverless, so that there the query costs are lower, so the game changes, but I would then see
60 00:10:25.920 ⇒ 00:10:30.969 Harsh Punjabi: You know, if there are any such options available between the two, compare the cost, and yeah.
61 00:10:30.970 ⇒ 00:10:34.879 Awaish Kumar: Do you know how the pricing model works between BigCary and Snowflake?
62 00:10:35.650 ⇒ 00:10:54.000 Harsh Punjabi: I mean, I can certainly do my research on that. In one of the older projects for solutioning, for Google Places API V2, I did… I did a cost audit, but, again, I… so they… you know, how much pool space it is consuming, I don’t remember the entire thing in detail, to be very honest with you.
63 00:10:54.000 ⇒ 00:10:58.239 Awaish Kumar: What are the optimization techniques in BigQuery?
64 00:10:58.920 ⇒ 00:11:03.819 Harsh Punjabi: So, normally when we, so, again, how, how many,
65 00:11:03.820 ⇒ 00:11:22.259 Harsh Punjabi: So, when we’re optimizing queries there, specifically using DataForm, which is, like, another Google product, how many query references are happening? How many full refreshes we are doing? Are we implementing SCD type 2 or not? Like, how, how, like, what is the incremental load in our data? All of that depends, yeah.
66 00:11:22.260 ⇒ 00:11:28.890 Awaish Kumar: I get it, like, these all concepts are important, but I’m trying to understand, like, at the end.
67 00:11:29.000 ⇒ 00:11:39.979 Awaish Kumar: when it reaches to BigQuery, right? What are the things that you need to do in terms while you are structuring it, right? Yeah. So it can help you improve the cost.
68 00:11:41.100 ⇒ 00:11:55.120 Harsh Punjabi: So, like, I’ll ask a follow-up question to clarify this a little bit. Are you asking from the perspective of how can I design my models to be low on cost, or how do I, you know, make sure people are running the queries in the right way so that the consumption costs are lower?
69 00:11:55.560 ⇒ 00:12:09.199 Awaish Kumar: I think it’s related, right? If my model is structured properly, and then people run queries on top of it, it will be the low cost, right? If your model is not properly architectured.
70 00:12:09.500 ⇒ 00:12:18.580 Awaish Kumar: then if they carry it, and the same way, it will be a lot of… like, the cost will be a lot high. So it all depends, like, the…
71 00:12:19.090 ⇒ 00:12:34.879 Awaish Kumar: on… on how you… like, the first step is how you structure it, right? Obviously. Then, second thing is how people use it. It’s… it’s… it’s one of the people, right? You just… you can just educate. You can go and train… train them, okay, these are the best practices. But before…
72 00:12:35.230 ⇒ 00:12:42.839 Awaish Kumar: Giving the, like, before going and sharing the best practices with the team, the first step is how we architect it.
73 00:12:43.440 ⇒ 00:13:03.030 Harsh Punjabi: So, normally what I take care of when I’m architecting this is I, if there are any cross-product, metrics, we try to put them into the star schema so that only the relevant fields are being pulled, it is not scanning the entire data, so that the cost remains lower. And, what, what one… another approach that we’ve done to keep costs down is.
74 00:13:03.030 ⇒ 00:13:11.760 Harsh Punjabi: We’ve kept, like, because there are different verticals within Lightspeed. I’m taking an example from here, and I’ll translate it to what I mean on a generic level.
75 00:13:11.760 ⇒ 00:13:25.179 Harsh Punjabi: But what we do is, the metrics which are common across all products live in a different data form project, and the hospital-specific live in a different project, and the ownership is clearly defined, like, at a model level. Which metric comes from which model is very clearly defined.
76 00:13:25.180 ⇒ 00:13:33.150 Harsh Punjabi: And, we break down the facts and dimensions in such a way that only the relevant dimensions are being pulled each time, so that the cost doesn’t get inflated.
77 00:13:33.230 ⇒ 00:13:56.270 Harsh Punjabi: And then the next layer to that is specifically within inside… within each model, right? Each specific dim table or each fact table, we try to make sure that, you know, how our refresh frequency works. Is it a full refresh? Do we do a one-time… like, how frequently are we doing a one-time historical backfill? How, are we keeping SCD type 2 where, you know, like, incremental data is being captured?
78 00:13:56.270 ⇒ 00:14:06.430 Harsh Punjabi: If the incremental data is being captured, what is the cost impact of that? Like, we obviously… that’s very situational on, you know, what specific case… use case we are handling, but depending on that use case, we kind of…
79 00:14:06.430 ⇒ 00:14:22.980 Harsh Punjabi: optimize for it. So, right now, I’m actually doing a migration, for, like, like, in the background, it’s not a high-priority project, but that’s something that, you know, has to be done. So, we are migrating, some, like, standard full refresh pipelines to SCD type 2 for the same reason.
80 00:14:22.980 ⇒ 00:14:28.200 Harsh Punjabi: That, you know, we need to have incremental backloads, and the primary key sanctity has to be maintained, yeah.
81 00:14:28.200 ⇒ 00:14:32.170 Awaish Kumar: Do you know the partitioning technique in BigQuery?
82 00:14:33.840 ⇒ 00:14:45.419 Harsh Punjabi: No, I primarily… I’m not really sure about it. I primarily worked with, you know, like, DataForm itself, which is very close to DBD, so there we take care of it within the models itself.
83 00:14:45.420 ⇒ 00:14:50.149 Awaish Kumar: Even if the DVD, like, that basically creates the table, right?
84 00:14:50.150 ⇒ 00:14:50.740 Harsh Punjabi: Yep.
85 00:14:50.940 ⇒ 00:15:07.779 Awaish Kumar: And if you don’t give it a proper configuration to create the partitioning on some columns, it will… it won’t create it, right? It will create a table without partitioning. And if you have a very big table, even if users are filtering it by days.
86 00:15:08.030 ⇒ 00:15:08.590 Harsh Punjabi: Nope.
87 00:15:08.590 ⇒ 00:15:11.629 Awaish Kumar: Your whole table will be scanned, because it’s not partitioned.
88 00:15:12.320 ⇒ 00:15:32.839 Harsh Punjabi: Yeah, so we define… so in the similar, just like dbt in the config block where you define the partition by XYZ columns, we do define that, but, like, I thought you were asking how does the partitioning logic in the background of BigQuery work? That I was a little uncertain about. But yeah, we obviously part… we index and partition the tables depending on the requirement of, each time, yeah.
89 00:15:33.360 ⇒ 00:15:35.149 Awaish Kumar: So does BigQuery have indexing?
90 00:15:37.940 ⇒ 00:15:47.379 Harsh Punjabi: It should. Honestly, to be very honest, it should have indexing. Any decent data warehouse should have indexing. But yeah, like, we configure that in data form itself.
91 00:15:47.380 ⇒ 00:15:50.489 Awaish Kumar: And they must have indexing, but there’s no…
92 00:15:50.880 ⇒ 00:15:55.559 Awaish Kumar: No way you can, like, provide a column for indexing.
93 00:15:55.820 ⇒ 00:15:59.670 Awaish Kumar: It might be handled inside
94 00:15:59.800 ⇒ 00:16:10.199 Awaish Kumar: And the, what you say, the back processes of BigQuery. There’s no feature like that for end users. We can only define partition by what call?
95 00:16:10.390 ⇒ 00:16:12.380 Awaish Kumar: Or we can define clustering.
96 00:16:12.770 ⇒ 00:16:13.400 Harsh Punjabi: Yep.
97 00:16:13.870 ⇒ 00:16:33.429 Harsh Punjabi: So, again, that’s what… DataForm is the backend. If BigQuery’s working with DataForm, which is another, like I mentioned, it’s a Google product. So indexing is possible in DataForm. We define a UDF for the indexing for a lot of pipelines, but yeah, if it is not in BigQuery, like, I was not aware about that. Like, in the front end of BigQuery, I was not aware about that.
98 00:16:33.880 ⇒ 00:16:36.130 Awaish Kumar: data from a big DBT, or…
99 00:16:36.390 ⇒ 00:16:42.639 Awaish Kumar: kind of users of BigQuery, right? They are on top of BigQuery, they don’t run inside of it, so…
100 00:16:43.090 ⇒ 00:16:49.249 Awaish Kumar: it can only use what we can use in UI. Okay, moving on…
101 00:16:49.460 ⇒ 00:17:02.489 Awaish Kumar: So, how do you make decisions between when to use Flettable, when to use the star schema, and when to use a snowflake schema?
102 00:17:03.730 ⇒ 00:17:07.410 Harsh Punjabi: Again, depends on the business problem. So…
103 00:17:07.480 ⇒ 00:17:14.969 Harsh Punjabi: My… my decision-making process is what is the most easy to… what is the most modular approach based on future changes?
104 00:17:14.970 ⇒ 00:17:38.629 Harsh Punjabi: So, if, if my… so, in cases where we have to keep on adding more and more products, and it’s a long table, not a wide table, we prefer a long table, and then we create a separate product dimension. You know, doesn’t matter, like, star and snowflake schema question depends on the layer of dimensioning that we have to do. So, within those two, depends on how many layers of dimensions we need to create.
105 00:17:38.630 ⇒ 00:17:41.980 Harsh Punjabi: to get to that decision. But, for the most part, it is…
106 00:17:42.160 ⇒ 00:18:01.789 Harsh Punjabi: the future use cases? How modular will it be? Will I have to break everything to rebuild, and rebuild it from scratch if a new product comes in or a new edition comes in? So the process I follow is, I first anticipate the future requirements. What could happen? Could a new product come in? Could new, good, you know, new derived metrics be created on top of the data that we are already working with?
107 00:18:01.790 ⇒ 00:18:07.359 Harsh Punjabi: And based on that, then I reverse engineer it and make a decision, and give a recommendation on how it should be done.
108 00:18:09.190 ⇒ 00:18:09.920 Awaish Kumar: Okay.
109 00:18:16.760 ⇒ 00:18:21.760 Awaish Kumar: Okay, so, yeah, you mentioned one thing about indexing, so…
110 00:18:21.870 ⇒ 00:18:24.269 Awaish Kumar: What is indexing, and how does it work?
111 00:18:25.140 ⇒ 00:18:36.589 Harsh Punjabi: So, basically, that’s your… indexing is kind of defining how your query will find points in your data, right? That’s basically what indexing is.
112 00:18:36.590 ⇒ 00:18:49.420 Harsh Punjabi: So, it… because if you… if you do not define an index and you start running queries, your costs will be high. It will scan entire tables and then, you know, maybe go at specific filter values, but if you’re able to define an index.
113 00:18:49.420 ⇒ 00:18:54.560 Harsh Punjabi: The, it becomes… it acts like an identifier, for your, you know.
114 00:18:55.960 ⇒ 00:18:56.530 Awaish Kumar: Yeah, you too.
115 00:18:56.940 ⇒ 00:19:03.960 Awaish Kumar: your… your query performance, right? But, yeah, how… how it does it, right? What is…
116 00:19:04.160 ⇒ 00:19:08.040 Awaish Kumar: Do you know anything about… How indexing work?
117 00:19:09.630 ⇒ 00:19:12.400 Harsh Punjabi: So, it’s, it’s…
118 00:19:12.960 ⇒ 00:19:23.909 Harsh Punjabi: So when you’re basically defining, that’s what, you know, you’re basically defining identifiers for your rows of records, so it takes that identifier as a point of reference, and only scans through specific
119 00:19:23.910 ⇒ 00:19:46.070 Harsh Punjabi: only it looks through the specific indexes that are needed, right? It will not, so, if I try to draw an analogy for it, if I have, like, 50 pieces of, you know, toys scattered on the floor, and I’m trying to find something, I don’t have to pick each one of them. If I know that blue toys are here and green toys are there, I can specifically go and pick the blue ones and the green ones instead of just
120 00:19:46.160 ⇒ 00:19:51.479 Harsh Punjabi: searching through the whole thing. That’s, that’s how, you know, indexing kind of helps.
121 00:19:51.480 ⇒ 00:19:56.570 Awaish Kumar: So basically, in the backend, like, it uses… B-trees.
122 00:19:56.900 ⇒ 00:19:59.310 Awaish Kumar: Yep. To handle that references.
123 00:20:00.330 ⇒ 00:20:05.679 Harsh Punjabi: Again, I’m not sure about that, but yeah, but that’s… I’ll look this up. I’ll be very honest with you.
124 00:20:05.680 ⇒ 00:20:10.380 Awaish Kumar: Yeah, a default, like, data structure that it uses.
125 00:20:10.670 ⇒ 00:20:15.160 Awaish Kumar: For example, in Postgres, To… to implement indexing, and…
126 00:20:15.780 ⇒ 00:20:20.310 Awaish Kumar: Okay, so, but then there… but different types of indexing, like, there are…
127 00:20:21.700 ⇒ 00:20:25.199 Awaish Kumar: Clustered and non-clustered indexing. Do you know the difference is?
128 00:20:26.640 ⇒ 00:20:34.640 Harsh Punjabi: I have… I’m familiar with the terms, I’ll be honest with you, I’m familiar with, like, clustered indexing, and I’m familiar with the terms, but,
129 00:20:34.810 ⇒ 00:20:45.400 Harsh Punjabi: you know, I do not specifically use it on a day-to-day basis, so I am not, you know, very fresh on that. I don’t remember those exactly.
130 00:20:45.400 ⇒ 00:20:52.020 Harsh Punjabi: But what I can anticipate, like, what it kind of means, I can take a stab at it based on my past understanding.
131 00:20:52.020 ⇒ 00:21:11.190 Harsh Punjabi: that, a clustered index would be a collection of indexes that it looks at together, and non-clustered would be individual indexes or discrete indexes that it will have to parse through. That’s what I can anticipate based on what my understanding is, but I’ll be very honest, I… it’s been a long time since I’ve, like, read about that, so, yeah.
132 00:21:11.620 ⇒ 00:21:19.190 Awaish Kumar: Yeah, no, no problem. Yeah, we are just, like, we just have left with 5-6 minutes, so I will leave this time for you to ask any questions.
133 00:21:19.770 ⇒ 00:21:36.940 Harsh Punjabi: Cool, so, yeah, I… I… when I was going through the Brainforge, you know, company page, and I was looking at, how things work, so I was curious to know, like, what, you know, there are… there are a bunch of services, like AI enablement, then data engineering, consulting, etc.
134 00:21:36.940 ⇒ 00:21:46.640 Harsh Punjabi: So, how does the team structure work? Is it, like, one specific team for specific clients, or is it, like, a team of people is doing everything for multiple clients? What type of products are being…
135 00:21:47.550 ⇒ 00:21:50.010 Awaish Kumar: We have a team for a specialization.
136 00:21:50.270 ⇒ 00:21:52.120 Awaish Kumar: T for a service.
137 00:21:52.770 ⇒ 00:21:56.169 Awaish Kumar: So if, if somebody, if a client needs AI,
138 00:21:56.830 ⇒ 00:22:02.219 Awaish Kumar: support, then we have an AI team that is going to give AI support.
139 00:22:02.510 ⇒ 00:22:11.420 Awaish Kumar: Right? If the same client needs data engineering support, then we have data engineers that are going to provide that support, and if…
140 00:22:11.460 ⇒ 00:22:23.029 Awaish Kumar: If the same client also needs data analytics engineering supports, like help with modeling, then we have a team for data… team of data analytics engineers that are going to provide
141 00:22:23.290 ⇒ 00:22:24.999 Awaish Kumar: That… that service.
142 00:22:25.240 ⇒ 00:22:28.700 Awaish Kumar: So, it is… Based on the service.
143 00:22:29.030 ⇒ 00:22:39.250 Awaish Kumar: And it is based on… based on the service, the team and the people are chosen, right? So it is going to be a service lead that defines
144 00:22:39.450 ⇒ 00:22:49.170 Awaish Kumar: That is going to define, okay, this client needs our services, and then he knows who to allocate within his service, who’s free, and have time for that.
145 00:22:49.490 ⇒ 00:22:52.020 Awaish Kumar: On the client side, we have a team of
146 00:22:52.800 ⇒ 00:23:00.739 Awaish Kumar: client success owners, right? That defines what this client needs, right? If the client needs 20 hours of
147 00:23:00.890 ⇒ 00:23:14.350 Awaish Kumar: AI work and 20 hours of data work needs 50 hours of AI work and 30 hours of data work. That is on… on client, like, customer success honor, like, the responsibility to define that.
148 00:23:14.940 ⇒ 00:23:19.839 Awaish Kumar: Right? And… and ask for… ask for service leads, right? I need…
149 00:23:19.940 ⇒ 00:23:23.319 Awaish Kumar: 20 hours from you, like, from your team, so, yeah.
150 00:23:24.490 ⇒ 00:23:35.970 Harsh Punjabi: That’s, that’s very, that’s interesting, that’s, that’s good to know. But that brings me to a follow-up question. So, because, teams are, by specialization of skill, right, of the task.
151 00:23:35.970 ⇒ 00:23:46.449 Harsh Punjabi: Normally, a lot of consulting companies are very industry-heavy, right? Some of them, like, the first company I worked with was pharma-heavy. The last one I… the last consulting company I worked with was…
152 00:23:47.050 ⇒ 00:23:47.770 Harsh Punjabi: Stay heavy.
153 00:23:48.330 ⇒ 00:23:54.760 Awaish Kumar: I get it, but we sell services not by industry. We sell services by the specialization.
154 00:23:54.760 ⇒ 00:24:07.189 Harsh Punjabi: My question is, is there a specific industry that, Brainforge… like, in the split of customers that Brainforge has, is there any specific industry which is dominant, or is it, like, a mix? Like, that was my question, basically.
155 00:24:07.190 ⇒ 00:24:22.270 Awaish Kumar: That’s… I understood that question, that’s why I’m answering. The way we target people is not based on the industry and the client we have. Obviously, we have some healthcare clients, we have some CPG clients, and…
156 00:24:23.130 ⇒ 00:24:30.640 Awaish Kumar: And that’s mainly why… the clients we have, right? We also have a client who… That…
157 00:24:31.150 ⇒ 00:24:34.599 Awaish Kumar: Client that manages the, for example, events.
158 00:24:34.720 ⇒ 00:24:37.450 Awaish Kumar: I, like, I mean, run the…
159 00:24:37.650 ⇒ 00:24:43.009 Awaish Kumar: events and sales the booths, right, for exhibitions and things like that. So…
160 00:24:43.980 ⇒ 00:24:50.380 Awaish Kumar: So, yeah, like, and then we have some clients which are maybe no longer with us. They were in industries like…
161 00:24:50.630 ⇒ 00:24:54.170 Awaish Kumar: Ecom and things like that, so… .
162 00:24:54.170 ⇒ 00:24:56.909 Harsh Punjabi: Mainly the US market, though. Like, mainly in the US.
163 00:24:56.910 ⇒ 00:25:02.529 Awaish Kumar: Mainly the clients are from United States, but not from a particular industry.
164 00:25:02.870 ⇒ 00:25:13.349 Harsh Punjabi: But that’s good. Actually, having a good mix of industries is always an advantage, because then you are not siloed. Like, personally speaking, my opinion, after having worked in consulting for over 5 years.
165 00:25:13.350 ⇒ 00:25:26.339 Harsh Punjabi: When consulting companies are very heavy on specific industries, then you’re kind of siloed into one specific type of work. But if there is a good mix, then, of course, there is a lot of scope for learning as well.
166 00:25:26.340 ⇒ 00:25:29.690 Awaish Kumar: comes from… that… that comes from the people, right?
167 00:25:29.690 ⇒ 00:25:30.200 Harsh Punjabi: Mmm.
168 00:25:30.200 ⇒ 00:25:34.789 Awaish Kumar: For example, the CSO, or the person who is trying to sell the service.
169 00:25:34.970 ⇒ 00:25:47.029 Awaish Kumar: Like, it is kind of by… kind of naturally comes from him. If he’s already working on 3 healthcare clients, and he has a lot of knowledge about how healthcare reporting works.
170 00:25:47.060 ⇒ 00:26:00.929 Awaish Kumar: It’s natural for him to hunt for someone who is in the same area, so he can quickly, like, he knows everything about healthcare reporting, the metrics, and how the dashboard looks like for the healthcare industry.
171 00:26:00.930 ⇒ 00:26:07.970 Awaish Kumar: And he can quickly go and show it and sell it. And it’s also natural that that client might come
172 00:26:08.200 ⇒ 00:26:12.130 Awaish Kumar: Like, convert, because it will see a…
173 00:26:12.510 ⇒ 00:26:15.549 Awaish Kumar: a real work, right, that we are already doing for somebody.
174 00:26:15.710 ⇒ 00:26:24.910 Awaish Kumar: And then, then, like, targeting someone where you haven’t… you don’t have anything real to show, and you just target based on your skills.
175 00:26:25.050 ⇒ 00:26:34.730 Awaish Kumar: So it’s… it’s… obviously, it kind of comes from that, like, when people are in the sales team, what are their,
176 00:26:35.040 ⇒ 00:26:38.049 Awaish Kumar: Like, the strong hand, where they can go and sell.
177 00:26:38.230 ⇒ 00:26:44.809 Awaish Kumar: But… but for us, like, we are trying to, come up with,
178 00:26:45.110 ⇒ 00:26:48.569 Awaish Kumar: services within our team, so we can tell the…
179 00:26:48.840 ⇒ 00:26:51.149 Awaish Kumar: Tell the sales team, okay, what to sell.
180 00:26:51.710 ⇒ 00:26:52.570 Awaish Kumar: Right?
181 00:26:52.730 ⇒ 00:27:00.570 Awaish Kumar: But, yeah, even if you try to do that, there are times when, obviously, the salesperson is the one who has to sail.
182 00:27:02.950 ⇒ 00:27:10.109 Harsh Punjabi: That’s helpful. I mean, I completely understand that. It’s a… it’s kind of a process-related thing as well. It makes sense.
183 00:27:10.140 ⇒ 00:27:29.249 Harsh Punjabi: I just have one more question, and that’s not brain force-related, but more of this role-related, so I wanted to take your perspective on, what do you think is, you know, the definition of success in this role? Or, you know, if, if things work out, then, what could I do to make sure that I’m successful in this role, or anybody who joins this role can be successful in this role?
184 00:27:29.520 ⇒ 00:27:33.589 Awaish Kumar: To be successful in this role is, first thing is.
185 00:27:33.740 ⇒ 00:27:39.400 Awaish Kumar: Communication and transparency. Since it’s a remote role, you have to be transparent.
186 00:27:39.570 ⇒ 00:27:46.200 Awaish Kumar: In terms of what, for example, what you’re working on, what… what is labeled, what is blocked?
187 00:27:46.350 ⇒ 00:27:50.619 Awaish Kumar: Pushing on things that are blogged and being active.
188 00:27:50.780 ⇒ 00:27:58.880 Awaish Kumar: In Slack, and… And then, yeah, like, and making sure, with your pod.
189 00:27:59.270 ⇒ 00:28:07.470 Awaish Kumar: that your clients are happy. That’s the main goal, right? So all the things that you are doing, you will be doing, like, okay, I’m done with tickets.
190 00:28:07.760 ⇒ 00:28:11.989 Awaish Kumar: That’s my update. Here’s the model, and…
191 00:28:12.550 ⇒ 00:28:18.660 Awaish Kumar: And I’m being active in Slack, answering your question from your analysts or CSOs.
192 00:28:18.950 ⇒ 00:28:24.719 Awaish Kumar: Basically, this is what we have to do to reach the goal of making sure that you are
193 00:28:24.870 ⇒ 00:28:28.510 Awaish Kumar: your CSO understands what you’re saying, and he can do…
194 00:28:29.100 ⇒ 00:28:32.259 Awaish Kumar: He can share that with the client, and the client is happy.
195 00:28:32.790 ⇒ 00:28:37.450 Awaish Kumar: So, the ultimate goal is the client needs to be happy with your work.
196 00:28:38.600 ⇒ 00:28:42.159 Harsh Punjabi: Makes sense, makes sense, provides a… provides a lot of perspective, yeah. Thank you, Avish.
197 00:28:42.160 ⇒ 00:28:42.730 Awaish Kumar: Yeah.
198 00:28:43.030 ⇒ 00:29:00.170 Awaish Kumar: So… but the… obviously, the kind of work that will land on your plate is… comes from your service lead. He’s going to define, for which clients are going to work, and how much, like, how your time goes where.
199 00:29:00.370 ⇒ 00:29:04.049 Awaish Kumar: And based on that, obviously, you will have your…
200 00:29:04.190 ⇒ 00:29:06.730 Awaish Kumar: workflow, and I’m going to deliver on that.
201 00:29:07.780 ⇒ 00:29:18.179 Awaish Kumar: We have a Slack, where, like, obviously I mentioned already, we have Linea, where we manage our tickets, but then we have Slack, where we communicate a lot with our team members.
202 00:29:18.300 ⇒ 00:29:27.070 Awaish Kumar: And… and then Zoom for meetings, and then we have Zoom clips, which we do, like, if you want to share
203 00:29:27.310 ⇒ 00:29:32.659 Awaish Kumar: Something which seems… which would… might be confusing for people, you can just…
204 00:29:32.790 ⇒ 00:29:37.620 Awaish Kumar: to promote a sent communication, instead of a meeting, you can just record two-minute
205 00:29:37.800 ⇒ 00:29:39.730 Awaish Kumar: Of your explanation and share it.
206 00:29:39.840 ⇒ 00:29:44.630 Awaish Kumar: And they are going to watch on their own time, so that’s how we manage.
207 00:29:45.800 ⇒ 00:29:54.459 Harsh Punjabi: That’s good as well, because I think, there’s a… there’s… mostly people have an overlap with the EST time zone, but everybody’s based out of different time zones, I suppose.
208 00:29:54.660 ⇒ 00:30:00.750 Awaish Kumar: Yeah, so it’s a global remote team, so everybody’s… We, we try to…
209 00:30:01.020 ⇒ 00:30:08.370 Awaish Kumar: Come up with some time, like, few hours, where everybody can meet each other, but…
210 00:30:09.120 ⇒ 00:30:13.869 Awaish Kumar: But yeah, people are in different time zones, and it’s easy if you share the…
211 00:30:14.040 ⇒ 00:30:19.719 Awaish Kumar: Clips, and try to communicate, like, communicate asynchronously as much as possible.
212 00:30:20.500 ⇒ 00:30:33.360 Harsh Punjabi: Absolutely. My current role is also remote, and this thing, you know, even though there’s not a time zone difference here, but, communication is certainly very different in remote roles than, you know, regular in-office roles. Yeah, absolutely.
213 00:30:33.360 ⇒ 00:30:37.470 Awaish Kumar: Yeah, thank you so much for today’s interview, and…
214 00:30:37.680 ⇒ 00:30:45.589 Awaish Kumar: after my… after I submit my feedback, or Kayla from our recruited team, like, she’s going to, come back.
215 00:30:45.760 ⇒ 00:30:47.490 Awaish Kumar: To you with the room.
216 00:30:48.170 ⇒ 00:30:48.910 Awaish Kumar: Okay.
217 00:30:49.330 ⇒ 00:30:52.160 Harsh Punjabi: Okay, thank you so much, Avish. Nice talking to you, really appreciate your time.