Apna College - Tech MasterClass with CTOs ｜ Angle One CTO & Shrad..
Category
😹
FunTranscript
00:00:00Hi everyone, today we are going to dive deep into the world of tech leadership with a true
00:00:04industry veteran.
00:00:06Today we have Jyoti Swaroop, who is currently serving as the CTO Chief Technology Officer
00:00:10at AngelOne, which is one of India's largest online stockbroker firm.
00:00:15Today we are going to cover a variety of things, including Jyoti's early days as a software
00:00:20engineer, his transition to e-commerce, then being a senior platform architect to building
00:00:25highly scalable products, and managing a massive team of 400 engineers as a CTO.
00:00:31We are going to dive deep into many things, including AngelOne's transition from a vendor
00:00:36to building its own in-house team of engineers.
00:00:39Jyoti shares how he managed this entire transformation, likening it to changing engines of a flight
00:00:45while the flight was still in the air.
00:00:47Today's conversation is also going to include, how do top tech leaders hire software engineers?
00:00:52What are the challenges to building highly scalable D2C products?
00:00:56What is the tech stack for AngelOne?
00:00:58How did they transition from a monolith architecture to a microservices architecture?
00:01:02Why did they choose AWS over Azure?
00:01:05And what are the different technologies that they decided to phase out?
00:01:08And lastly, we'll talk about the primary responsibilities of being a CTO, the compensation for a CTO,
00:01:13the future and impact of AI in fintech, and what does it mean for the industry?
00:01:18So whether we are a budding engineer or an aspiring CTO, this entire conversation is
00:01:23going to be no less than a masterclass.
00:01:25We are going to get a lot of insights, we are going to get a lot of advice, we are going
00:01:28to get a lot of perspective on how can we navigate our tech career better.
00:01:36So today we are with Jyoti, who is currently the CTO of AngelOne.
00:01:41So Jyoti, I'll start with the most obvious question.
00:01:43What has your journey been like in the technology sector?
00:01:47And what are the events that led to you becoming the CTO at AngelOne?
00:01:51So yeah, so I liked coding from early time, basically crafting out, you know, new experiences
00:01:59out of thin air, right, was something was very fascinating for me.
00:02:03So I used to dabble around in that.
00:02:06I started off my career, you know, very specifically getting into places where I could build end
00:02:14to end products, right, full stack, if you will, the back end, the front end, all the
00:02:20pipelines, etc.
00:02:21You know, and I didn't go for brand and stuff like that, or bigger companies, where I would
00:02:28be solving point problems, right.
00:02:30So I worked on things like aerospace control systems, real time operating systems, I even
00:02:39worked on the Windows kernel at Microsoft, right.
00:02:42I dabbled in that for about 10 years, before I got into e commerce and distributed systems.
00:02:51So I've done three variants of e commerce, retail, travel, and now fintech.
00:03:00On your last part, how did I get into AngelOne, see when I, so I was always little apprehensive
00:03:07about stock market for a longer longest time, more of a SIP kar ke bhool jao type person,
00:03:13right.
00:03:14But during the COVID time, you know, I saw a lot of my friends and people in my team
00:03:20actually be very active on the stock market.
00:03:24I remember someone saying, why don't you buy Hawkins, I thought the guy was talking about
00:03:29a pressure cooker.
00:03:30Then I realized he was talking about buying Hawkins share price.
00:03:35So I got interested on why people like doing it, and what's the mechanics below, you know,
00:03:41how does it work.
00:03:43And I found the domain to be very interesting, right.
00:03:46So when AngelOne came up with this offer, and they wanted to basically rejig the old
00:03:54product in tech stack, I was very happy to be in the driving seat.
00:03:59So you worked at a variety of companies, you worked at Microsoft, then you went to Walmart
00:04:03and Goibibo.
00:04:04And now you're at AngelOne.
00:04:06So which one has been your favorite company and why?
00:04:09That's a very tough question.
00:04:12Honestly, I think the most fulfilling has been AngelOne, because, you know, the problem
00:04:19space was so diverse.
00:04:21You know, and unlike travel and e commerce, you depend a lot on partners, right.
00:04:28For example, in airlines, we don't want inventory, like, you know, people like Indigo, Spicejet
00:04:33and other folks do, right.
00:04:35So a lot of things we need to rely on it.
00:04:37So that kind of gives you a lot of constraints.
00:04:39Similarly, in retail, there are a lot of constraints, right, and you need to work within that.
00:04:46In the fintech space, I would say the constraints are much lesser.
00:04:51So what you can build as a product is constrained by your creativity and skills, right.
00:04:57So that was really amazing.
00:04:58I also got to build a pretty solid team pretty much from ground up.
00:05:04So I think everything else considered, this has been the most fulfilling part of the career
00:05:10so far.
00:05:11Okay, nice.
00:05:12So throughout your experiences, what were your key lessons that helped you in your current
00:05:17role when you're working as a CTO?
00:05:20I think I'll talk about some technical lessons and some probably leadership lessons for like
00:05:27of a better word, right.
00:05:28So technical lessons, I think the first thing is keep things simple, right.
00:05:32I think many engineers tend to prematurely optimize or prematurely abstract out things,
00:05:39right.
00:05:40They start building a platform and then think about what they're going to be, what's going
00:05:45to happen when they hit high scale, etc.
00:05:49And then it leads to more complications than necessary, right, because it takes time for
00:05:54people to even understand the problem, right.
00:05:58And yet you're building some stuff out for assuming that you're going to be very successful
00:06:03with the same solution, right.
00:06:05So the first lesson is don't prematurely optimize or abstract out, right.
00:06:10First solve the problem, get it right, then get it scalable, right.
00:06:16The other, you know, when we build software today, right, it's usually utilizing different
00:06:22components together.
00:06:24We don't write software in isolation, your code will make some API calls, it will be
00:06:28housed in some cloud solution, it will use probably a managed database, etc, etc, right.
00:06:35They are all things outside of control, right.
00:06:38So one another key lesson is healthy distrust of anything which you don't control, right.
00:06:44So it's not a question of if something will fail, it's a question of when will it fail.
00:06:50And you know, like, what would you do to handle that failure?
00:06:54If you plan things like that, you can build more reliable stuff, right, rather than assuming
00:07:00that this API will always work.
00:07:03And there's a structured process on how you can do this, right.
00:07:07For example, if you're making an API call, which is most people when they're building
00:07:11apps and you know, they make API calls and to craft some new experience, right.
00:07:16So when you're making an API call, you need to ask yourself, how much time am I willing
00:07:20to wait?
00:07:21Right?
00:07:22Because you don't know the other system has failed or not, you might be waiting forever.
00:07:28So you need to see I need to give it a budget.
00:07:30So let's say I'll give it one second for the API to respond, right.
00:07:34And then you need to think whether you need to add a retry, will I retry?
00:07:38How many times would I retry?
00:07:40Now, once you do the maths, then your guarantees become evident.
00:07:46One second timeout, let's say you want to retry twice, then your bare minimum time that
00:07:52you can promise your customer is two seconds, right.
00:07:55So generally, you work the other way around, saying what do you want to promise your customers,
00:08:00and then divide up the budget, right.
00:08:02So having this structured thought process and healthy distrust, I think is very helpful
00:08:09in building, building things.
00:08:11The third lesson I would say is be lazy.
00:08:16You know, think a lot more before, I think it ties into the first point only, right?
00:08:22I mean, think more.
00:08:25And you know, before you kind of get into a lot of the nitty gritties of execution,
00:08:30right, what I've seen is people, engineers who are lazy, generally build better solutions,
00:08:36because they don't want to do copy paste and a lot of other things, right, or manual work.
00:08:40So they'll figure out ways to get things done by tooling, etc, which generally is much
00:08:47more scalable, right.
00:08:50In terms of the team, what I try to do is, you know, there's a fine balance between,
00:08:56you know, allowing a lot of diverse opinions, because engineers are opinionated, very varied.
00:09:01So everyone would have a different way of doing the same thing.
00:09:05And you don't want to eco room, right, you don't want a room where, you know, some, let's
00:09:09say a senior engineer says something and everybody says, okay, right, we want to have a dialogue.
00:09:15So you know, and that, that comes out by enabling, like, you know, psychological safety for engineers.
00:09:21So even a new person can actually challenge something, you know, or ask some questions.
00:09:27So you need to enable that environment.
00:09:29At the same time, as a company, we need to make decisions, right, and we need to get
00:09:35out products, right?
00:09:38Because every opportunity has a time window, it's not like this, you know, this opportunity
00:09:42is going to be lasting as forever, right?
00:09:45So how do you balance both, right?
00:09:47How do you have that dialogue and difference of opinions, but still kind of disagree, but
00:09:54commit within a time frame, I think that is a art more than science, right?
00:10:00But I sincerely try to enable it for the teams.
00:10:07I also like to keep teams lean, right, without too much, you know, so that engineers and
00:10:13product designs kind of work together.
00:10:15So if you don't mind, like, what is the current size of the tech team and the prod team at
00:10:19Angel right now?
00:10:21So we have about 400 people in tech, and I think about 150 people in product, right,
00:10:28which is pretty big, right?
00:10:30But if you look at the, we have structured them in pods, right?
00:10:35So if you look at a specific team, they are still lean, and you know, so, and they connect
00:10:40to each other, rather than going up in a hierarchy and connecting down, right?
00:10:44They talk to each other, they take most decisions locally, right?
00:10:50So that's, that's what we try to do.
00:10:54Okay.
00:10:55So, you joined AngelOne in 2021.
00:10:59So how would you compare the tech team of AngelOne in 2021 to today?
00:11:03What are the changes that has been?
00:11:06Yeah, see, I mean, very honestly, AngelOne started as a broking company, right?
00:11:12It was not a, you know, and, you know, but that said, our founder had always believed
00:11:18in tech, and, you know, tried to get a very tech centric solutions to product.
00:11:26And I read about his story, and how he started using walkie talkies when he started with
00:11:30the firm.
00:11:31Yes, exactly.
00:11:32Right?
00:11:33Walkie talkies in the, in the ring, right?
00:11:34So that they could give information back to customers and also always looking at solutions,
00:11:40right?
00:11:41When I joined, a large part of the team was vendor driven, where some specific team was
00:11:49handling the apps, and we used to kind of outsource it to those vendors.
00:11:55And similarly, there are a lot of contractors for back end systems, etc., right?
00:11:59And then we had a few core engineers, and this was during COVID.
00:12:02So we had high, you know, the engineers were spread across a bunch of different cities
00:12:08in India, right?
00:12:09So the difference between there and now I think, I think it's a good blend of the people
00:12:17who were here for a long time and had good domain knowledge and knew what works, what
00:12:23didn't work in financial ecosystem, and people who came in from outside, who had some best
00:12:31practices on, you know, what works in distributed systems, or building B2C apps, you know, what
00:12:38work what what appeals to customers and stuff like that.
00:12:42But we're not necessarily fully conversant with the finance, in tech domain.
00:12:47Okay.
00:12:48Right?
00:12:49So I think now people have exchanged ideas and skills.
00:12:51So it's a good marriage.
00:12:54Nice.
00:12:55So talking about newcomers in the team.
00:12:58So what do you look into a new member when you are hiring somebody new for your tech
00:13:02team at Angel1?
00:13:03Yeah.
00:13:04So we, so first is computer science fundamentals, right?
00:13:07And we stick to fundamentals, we don't ask weird DB questions and stuff like that.
00:13:13Problem solving skills.
00:13:15And again, it depends on what level you are coming in.
00:13:17So if you're a fresher, we don't expect you to be very, you know, we don't ask very involved
00:13:22system design questions, right?
00:13:25We look at how you structure your code, how do you, whether you ask questions, right?
00:13:29Or whether you just run with whatever assumptions that you made from the question.
00:13:33Right?
00:13:34So we look at that how you solve the problem, the actual final code, we look at it, but
00:13:39the process of arriving at the code is more, more important than the actual code that you
00:13:43write within in your coding interview.
00:13:46Right?
00:13:47So problem solving skills.
00:13:49And the other part would be curiosity.
00:13:51If you are asking questions, you know, if you have shown initiative in understanding
00:13:58things which were not required for you, whether in your college time, things like that.
00:14:03Adaptability, you know, if you are under pressure, like how do you react, right?
00:14:10And whether you can collaborate with others.
00:14:12So we don't ask domain questions, even for very senior folks.
00:14:17Right?
00:14:18We don't ask very domain specific questions, right?
00:14:21Like, what, how do you do this in Java and stuff like that, we don't ask that, right?
00:14:26For senior engineers, we look at specific examples in their lives, in their professional
00:14:33lives, they have shown specific attributes.
00:14:36So collaboration for a, you know, a fresher, I mean, we use the judgment of the interview
00:14:43because it is very difficult to make it out, right?
00:14:44Yes.
00:14:45But once we are hiring someone laterally at a very senior level, we ask for very specific
00:14:49examples on, on let's say collaboration, right?
00:14:54Can you demonstrate how, what kind of, you know, a situation where collaboration was
00:15:01required?
00:15:02What do you do?
00:15:04And also a time when you did not feel it happening.
00:15:09Even in a place where others were not collaborating with you, what did you do?
00:15:14Things like that.
00:15:15Right?
00:15:16We also sometimes ask open ended questions, like, I don't know if, like in Google, called
00:15:23GCM.
00:15:24Googling this.
00:15:25Yeah.
00:15:26The cognitive abilities.
00:15:27I mean, just to, again, to see your thought process, because the process is very important
00:15:31for us.
00:15:32Right?
00:15:33So, how you clarify things and how you think through a problem statement.
00:15:38So, can you please give a few examples of what kind of open ended questions you ask?
00:15:43Okay.
00:15:44Like, okay.
00:15:45So, one example, you know, like, again, it depends on the levels, right?
00:15:50So, but I would, if someone was a fresher, right?
00:15:52So, I would ask something like, how would you, let's say, you know, colonize Mars?
00:15:59Right?
00:16:00So, so, and then we look at the, as I said, right, the process, I mean, what is probably
00:16:10not a great idea is going on a rant, saying, I will do this, I will do this, I will email
00:16:15Elon Musk and stuff like that.
00:16:18Right?
00:16:19Because that means you're not asked relevant questions.
00:16:21Right?
00:16:22I mean, do you want to colonize Mars for the entire Earth or just for a few people?
00:16:29What time frame are you looking at?
00:16:31Right?
00:16:32Is there a budget?
00:16:33Right?
00:16:34So, so, those kind of questions should come in your mind, right?
00:16:37And kind of, you should build a problem space around at least a high level, you know, before
00:16:43going, going into things.
00:16:45Right?
00:16:46As I said, when you are just out of college, we have a larger room for, you know, like,
00:16:54not doing, not asking questions and going about this, but we still look at how you think
00:16:59through.
00:17:00The senior, more senior you are, if you, if you, if, if you have a bias for making assumptions,
00:17:08it's a very serious red flag.
00:17:10Okay.
00:17:11Right?
00:17:12Okay.
00:17:13So, you said that before 2021, there were vendors involved and then you kind of started
00:17:18with the in-house team.
00:17:20So, did you build the team from scratch?
00:17:22And did you also build the technology from scratch or was some code taken from the vendors?
00:17:28Yeah.
00:17:29So, it's, I think we, we changed a lot, both in terms of the code and other things and
00:17:35the DNA.
00:17:36Right?
00:17:37But it cannot be a zero to one transition or overnight transition.
00:17:39Right?
00:17:40So, we had to change the engines while the flight was in the air.
00:17:44Okay.
00:17:45Right?
00:17:46And we had to do it bit by bit by bit and iterate on top of it.
00:17:49Honestly, the first six months we're spending, we're spent in, ruggedizing the current things,
00:17:57whatever it was, business was growing.
00:17:59Right?
00:18:00I mean, we were growing at a very fast pace.
00:18:04Right?
00:18:05And we didn't have time to change everything.
00:18:08And I also didn't know what, what needed to be changed.
00:18:11Right?
00:18:12So, first six months, we just spent time making sure things were stable.
00:18:18You know, apps didn't crash, you know, we had monitoring, alerting, things like that.
00:18:25Right?
00:18:26So, that enabled us to understand systems better, and then plan for what needed to change.
00:18:31Like, what are the recipes, what tech stack we should build in, things like that.
00:18:36Yeah, in terms of people also, as I said, we had some people here, besides the vendors,
00:18:43like some in-house engineers, and we also hired a lot from other, like, I would say
00:18:51B2C companies.
00:18:52Right?
00:18:54So, we have hired people from Amazon, Walmart, Microsoft, you know, Uber, and stuff, places
00:19:00like those as well.
00:19:02Right?
00:19:03And they have a taste of executing on requirements, building for scale, right?
00:19:10Things like that.
00:19:11So, we wanted that DNA.
00:19:13One more very obvious question.
00:19:15So, as a CTO, what are your primary responsibilities right now?
00:19:20So, my primary responsibilities, first and foremost, is building scalable teams, empowering
00:19:27them to execute, aligning them with the company's vision, and making sure our tech is robust
00:19:35and scalable.
00:19:37And innovation, right?
00:19:39Helping the team with the overall company, right?
00:19:44And the last part, I run another small team called Labs, where we look at two to five
00:19:50year horizon of how do we build stuff for customers at that time scale level.
00:20:00Because in BA, you generally do delta by delta by delta, right?
00:20:05And sometimes, you know, it helps you, you know, you're dancing on the dance floor.
00:20:09Sometimes it helps if you can go up to the first floor and see what's happening.
00:20:13Right?
00:20:14So, that's what we do in the Labs team.
00:20:16Right?
00:20:17Yeah.
00:20:18So, it is both technical and managerial swim lanes.
00:20:21And I try to drive in both lanes, obviously, different times.
00:20:25So, I get involved in major architecture and design reviews of new things that we are building.
00:20:33I look at, you know, I look at, I also am very keen in attending all incident calls.
00:20:40So, whenever something happens or there's abnormality detected, right?
00:20:44I generally join the incident calls, even if I have scheduled meetings, right?
00:20:49And I may not talk in the meeting, but I listen and I see what's happening.
00:20:53Right?
00:20:54Those meetings give very good signals.
00:20:59It tells you, for example, how our customers are doing with a specific new feature or a
00:21:04new system.
00:21:05Right?
00:21:06So, how our team is able to respond.
00:21:10How robust.
00:21:11Yeah.
00:21:12How robust it is.
00:21:13Do people are able to debug?
00:21:14That's the first thing I look at.
00:21:16Right?
00:21:17Yes.
00:21:18Many times, sometimes I've seen situations where people don't know what to do.
00:21:21Because, you know, we see, let's say, convergence going down in a page.
00:21:26Right?
00:21:27We don't know what happened.
00:21:28Right?
00:21:29And then people spend some time figuring out what happened.
00:21:32And so, I see whether we have the right telemetry, we are able to debug what happened.
00:21:37Are we able to, how fast are we able to correct the situation?
00:21:42Right?
00:21:43And that is a spectrum.
00:21:44Right?
00:21:45One is you, let's say, this is a new thing that you released, we release about five times
00:21:50in a day.
00:21:51Okay.
00:21:52Across.
00:21:53Right?
00:21:54So, one easy thing to do is, actually, one great thing to do is roll back.
00:21:56Right?
00:21:57Whatever you did yesterday night.
00:21:58Right?
00:21:59Roll it back.
00:22:00And then debug.
00:22:01Right?
00:22:02So that you don't keep the customers impacted.
00:22:04Right?
00:22:07So, I see if that's happening.
00:22:08Right?
00:22:09I see how people behave under pressure.
00:22:11Right?
00:22:12How does the team lead behave?
00:22:14You know.
00:22:15Again, a red flag is someone saying who's responsible.
00:22:18Right?
00:22:19As the leader, team lead, that person is responsible.
00:22:23Right?
00:22:24So, a more constructive discussion would be, what do we do now?
00:22:28Do we have the data?
00:22:30So I see how people are navigating through things, how people are responding.
00:22:33Right?
00:22:34Incident calls is something which I really like to join.
00:22:38Not that I like incidents.
00:22:39You know.
00:22:40And RCA meetings.
00:22:41Right?
00:22:42RCA meetings are also very, very instructive.
00:22:46You know, like, I look at how people analyzed something which happened, whether they thought
00:22:52through about why it happened in real depth, and what are the plans they have.
00:22:58Okay.
00:22:59Right?
00:23:00I mean, this is the fun part of the work.
00:23:02Then there's obviously, you know, the other part of the work where I said I need to make
00:23:08sure people are empowered, hiring, and overseeing the overall tech people strategy for the company.
00:23:19So since you've worked with a lot of highly scalable D2C products, did you face any major
00:23:25challenges or did you, like, build something which stands out in your memory, which you
00:23:31would like to share?
00:23:32Okay.
00:23:33So, so, let me give a idea about the mechanics first on how, let's say, stock broking work.
00:23:42Stock broking is the most fun and most complicated part.
00:23:45Right?
00:23:46So, basically, it's a marketplace.
00:23:49Right?
00:23:50There are buyers and sellers of specific stocks.
00:23:53Right?
00:23:54And when does a transaction happen?
00:23:55You know, when does a buy and sell get matched?
00:23:58Right?
00:23:59So, a buy and a sell get matched when they are close together in price and time.
00:24:04Right?
00:24:05So, that means, if I am buying and you are selling, the prices have to be close together
00:24:10and we need to be, you need to send an order and I need to send an order about the same
00:24:16time.
00:24:17So, there's a queue.
00:24:18And then there's a matching algo at the queue, at the exchange.
00:24:21Right?
00:24:22Now, why would you and I sell at the same time?
00:24:25Right?
00:24:26Because we have some data, we keep looking at, let's say, price action.
00:24:30Right?
00:24:31How the price of a stock has moved.
00:24:32Right?
00:24:33And I feel this is the right time to buy.
00:24:34You feel it's the right time to sell.
00:24:35Right?
00:24:36So, we look at data.
00:24:37Right?
00:24:38And that data also keeps changing a lot.
00:24:40And also, if I see the opportunity or a customer sees the opportunity, you want to be fast.
00:24:46Because, as I said, if they have to be close in time, if, you know, if I want to buy, if
00:24:52I want to sell and you want to buy, if I'm late, then you would buy someone else's stock.
00:24:57Right?
00:24:58And I would miss out.
00:24:59So, the key problem is, how do we ship data from the exchange, where matching is happening,
00:25:05to the apps?
00:25:06Right?
00:25:07We have the 5 million odd active apps which are connected.
00:25:10How fast we can ship that market data to all these apps?
00:25:13In your phone, for example.
00:25:15Right?
00:25:16And how fast can we ship your orders back to the exchange?
00:25:20Right?
00:25:21So, this is the problem.
00:25:23And then we need to do this over mobile networks, while you're driving a car and going through
00:25:27a tunnel.
00:25:28Right?
00:25:29As servers fail, you know, and a lot of other things happen.
00:25:34Right?
00:25:35The path between the customers and the exchange is very complicated, very diverse.
00:25:39Right?
00:25:40You might be on a carrier, I might be on a different carrier and stuff like that.
00:25:44So, this was the problem to solve.
00:25:47And it involved building a lot of different solutions to help us build a scalable pipe,
00:25:55a real-time fast pipe, over all these things which are there.
00:26:00Next question, like it's personally my question.
00:26:04How is the hiring process for CTOs or for senior level engineers, maybe a team lead,
00:26:11different from other engineers in the industry?
00:26:14Yeah.
00:26:15So, when you are a senior engineer or a senior leader, right, your job is more of an enabler
00:26:22rather than a doer.
00:26:23Right?
00:26:25It's more of a resource multiplier.
00:26:26Right?
00:26:27Why would, when you're a senior engineer, you need to, what that means is, you need
00:26:30to make people who are along with you more productive, more successful.
00:26:35Right?
00:26:36When you are reasonably early in your career, you are more of a doer, and you're, right,
00:26:43so that is what we look at.
00:26:44And that's the differentiator, actually, on how we evaluate, right?
00:26:48So, when you are, let's say, at a senior engineer level, right, we look, we over-index
00:26:55more on things like system design, for example, right?
00:26:59How crisp are you?
00:27:00I mean, do you ask questions first, what we talked about, right?
00:27:03Do you ask clarifying questions?
00:27:05Are you able to crystallize a design which is neither over-solving, but also does not
00:27:13lose out any of the problems, the key aspect of the problem statement, right?
00:27:19Do you look at all aspects of the solution, right?
00:27:21Not just, okay, I will, you know, let's say I want to search for hotels in an area, right?
00:27:28Do you just build something and make choice?
00:27:30You use some database because you've read it in the blog, or do you go have a structured
00:27:36process of analyzing what to use, right?
00:27:41Do you look at instrumentation, how to go to production, you know, logging and stuff
00:27:47like that, right?
00:27:48Do you consider all that when you go live, right?
00:27:50So, we over-index on that, right?
00:27:53And less on, let's say, algo design or solving a, how you, how fast you're coded, etc., right?
00:28:03For people who are younger in their journey, careers, we look at more about whether, how
00:28:11thorough is your code, right?
00:28:13Whether you have written clean interfaces, stuff like that.
00:28:18Interfaces are very important, you know, because they are your contract with the outside world,
00:28:23right?
00:28:24Whatever you build underlying, underneath the interfaces, you can change on your own,
00:28:29right?
00:28:30But if you give me a contract, and then I have to, let's say, I am the front-end person,
00:28:34and I have to write, you know, call your API, then we are kind of bound on that API.
00:28:41And if you make some change there, then it's a very, it's a big change for everyone, right?
00:28:46So, we look at how well you design APIs and interfaces and code and stuff like that.
00:28:51So, those are the, I would say, the differences.
00:28:56Computer science fundamentals and many things are the same, right?
00:28:59So, those things stay the same.
00:29:01So, do they still ask you computer science fundamentals when you get hired as a CTO?
00:29:07I mean, they ask me the thought process of why choices were made, right?
00:29:17And then you need to articulate a clear thought process there.
00:29:21The fundamentals are useful even at that level, right?
00:29:29For example, how did you scale a business from, on a hockey stick curve, right?
00:29:35Hockey stick curve is when the business is growing, right?
00:29:37And the tech has to keep up, you know, or enable that growth.
00:29:43What are the choices that you make, right?
00:29:45So, then once you go down that, right, or generally one of the things I like to ask
00:29:50my, someone whom I am hiring at a senior level is, describe an incident where you drove the
00:29:57incident car.
00:29:58What questions did you ask?
00:29:59Right, if you are an application developer, do you understand system software networking,
00:30:06things like that, right?
00:30:07How do you identify bottlenecks in your code?
00:30:11So, yeah, fundamentals are still come into play, but nobody will ask you a thing like
00:30:18what is the.
00:30:19Pagination.
00:30:20Yeah, pagination or what is the, you know, like, I do not think anybody asks questions
00:30:23like what is the complexity of a map lookup, right?
00:30:27Nobody, nobody will ask you that.
00:30:30People will see whether you employ maps in the right place at the right time.
00:30:34Next question would be, what is the current technology stack of Angel 1?
00:30:39And throughout the process of building tech, how did you decide which technologies you
00:30:43had to adopt and which technologies you had to phase out?
00:30:47Okay.
00:30:48So, so we have a microservices architecture, right?
00:30:51So, there are multiple services which power different experiences on that.
00:30:55Did you start with microservices or did you started with monoliths?
00:30:59Yeah, we had a monolith, right?
00:31:00And that is how generally most people start, do not over optimize or over platform, right?
00:31:04As I said.
00:31:05So, we started, I mean, it was a monolith for the longest time.
00:31:11But when we saw the scale and we wanted to independently execute on different features
00:31:16and stuff like that, we went to a, I mean, we migrated towards a microservices architecture,
00:31:22right?
00:31:23So, that means each team is independently working on code in their own repos and deploying
00:31:29stuff to prod and testing and everything in between, right?
00:31:34So, this is the, you know, so we also saw a few design patterns which are very important
00:31:41and we made them into a toolbox for people to employ, right?
00:31:45So, one such example is the pattern called CQRS, right, where you separate the write
00:31:51and the read path, right?
00:31:52So, for example, as I said, we need to be fast, right?
00:31:56So, the write path has to be very, very lean, where write is basically order placement,
00:32:01right?
00:32:02The read here might be, let us say, you looking at your portfolio.
00:32:04There are other reads like positions, I am not talking about that, but a portfolio is
00:32:08a pretty good delayed read.
00:32:10So, you buy stock X, you want to see stock X, right?
00:32:14Those paths are handled differently, right?
00:32:18And then we can do what is right for the write path and what is the correct thing to
00:32:21do for the read path, so stuff like that, right?
00:32:26These services are deployed into types of data centers, private and cloud, right, or
00:32:33the core trading happens in private data centers.
00:32:36Everything happens, everything else happens on a public cloud, right?
00:32:40So, this is the core deployment and when the apps are there, the apps talks to different
00:32:46DNS names.
00:32:47So, when we are doing order placement, the core journeys, from the app, you directly
00:32:52get connected to the private data center, there is no router in between, right?
00:32:57And also, let us say portfolio is another example of what we have deployed in the cloud.
00:33:03So, if you make any portfolio API calls, the app talks directly to AWS, there is no middleman.
00:33:12In terms of the backend, we use Go, right?
00:33:16And Go is one of my, has been one of my favorite languages for a long time, right?
00:33:22And as I said, we had a diverse team, we had people who were incumbent and people are coming
00:33:28in from different companies, some from Java world, some from Python world.
00:33:33So, Go is something that you can get used to very quickly, right?
00:33:38And it is very simple to write code, right?
00:33:40So, that is why, you know, and also it is very, very efficient.
00:33:44So, our traffic scales up literally at a big cliff, right?
00:33:50So, 914 is hardly any traffic, 915 is a steep wall, right?
00:33:55And Go is great at that, right?
00:33:57And there is a lot of concurrent things that we do, as you can imagine, right?
00:34:00Multiple people are placing orders, so everything is happening at the same time.
00:34:04So, if you look at Go, the way you code concurrent programming, like handling requests, for example,
00:34:12you can do it in a much easier way than some other languages, let us say like Java, for
00:34:19example, right?
00:34:20It is very native to Go in coding parallel programming, in concurrent programming, right?
00:34:29And it is also very efficient.
00:34:31So, I think we are familiar with threads.
00:34:33So, every request is handled on different threads, right?
00:34:37So, in Go, a thread size is 2k, right, a standard and it grows and shrinks.
00:34:42So, if you have more, if you call a function which has a fetter object, the stack increases,
00:34:48because, you know, the function call is maintained at the stack, right?
00:34:52Whereas, if you use something like Java, right, it uses the OS thread, whatever the OS provides,
00:34:58the same thread is available in the app, which is about, if I remember right, in Linux,
00:35:03it is about 16 MB.
00:35:05Okay.
00:35:06Now, what does this mean?
00:35:07We have a 4 core machine, 4 gig, let us say RAM in a server, right?
00:35:12A single machine can host about 2 million goroutines or threads versus, let us say,
00:35:18I think it will be about 512k threads, right?
00:35:22So, there is a huge difference in how much firepower you can pack on a hardware, right,
00:35:31without losing any of the latency guarantees and stuff like that, right?
00:35:37So, that makes things much more efficient, right?
00:35:40You can code in a much easier way, you can express yourself in a much easier way, it
00:35:45is very cost efficient, it is very powerful, that is why we chose Go.
00:35:50On the front end side, we use native languages as much as possible, the native stack, because
00:35:55we need that performance, we need, right?
00:35:59And for some use cases where, you know, let us say, performance is not a main criteria,
00:36:06but we want to go live fast.
00:36:09And so, we use, look at cross platform UI stacks, pretty much it is a web view.
00:36:14And we have built some secret sauce on top of the web view.
00:36:18So, you build it once and it is available on all platforms.
00:36:21On the front end side, we use solid, which is, I mean, many people have heard about React,
00:36:26right?
00:36:27So, it is a similar stack in JavaScript.
00:36:32The difference is solid is much more reactive, right?
00:36:35You know, in terms of what reactivity is, reactivity is something happens, how does
00:36:42the UI update itself, right?
00:36:44So, there is a, like an observer, there are multiple observers.
00:36:48And then, you know, there is like the event bus, which is happening, event bus could be,
00:36:52let us say, a price tick coming in, a price has changed from, let us say, 2 to 5 or whatever,
00:36:58right?
00:36:59So, now, different panes need to react to the change and update, they update themselves.
00:37:05If you are slow in that, the customer will not have a good experience, because something
00:37:08will tick here, something will not tick there and stuff like that, right?
00:37:12So, we want everyone to react as fast as possible, because solid is great for that, right?
00:37:17So, and that is why we use solid in the front end.
00:37:21For databases, we use standard ACID databases.
00:37:25I am not a big fan of, you know, I would say, it is not a big fan, I mean, I think, like,
00:37:32they are, the MongoDBs of the world have their own use cases, but they are also heavily abused,
00:37:38right?
00:37:40So, the ACID semantics of a database, even though it is very old school, it actually
00:37:45makes for a much better programming model.
00:37:48And the SQL engines today are, so, you know, they scale pretty well.
00:37:52So, we use databases, you know, we try to avoid, or rather we try to stick to the ACID
00:38:02compliant databases as much as possible.
00:38:04Did you phase out any technologies that you used to use?
00:38:08Yes.
00:38:10So, you know, and a lot of the code, let us say, on the back office side, which is doing
00:38:16all your billing and a lot of other things, so, we used to have a lot of code in what
00:38:21is called stored procedures.
00:38:22So, and it is a very easy way to start implementing certain logic, things start very small.
00:38:30I want to do some calculation of, let us say, GST, right?
00:38:34It is very easy to plonk a very small snippet of code and say, okay, this will give me GST,
00:38:40right?
00:38:41But once you start that, it often gets abused and you start dumping more and more code into
00:38:48that, till at some point in time, you do not even recognize, you yourself do not recognize
00:38:52what you wrote.
00:38:53Right?
00:38:54So, we moved, we kind of moved that out, you know, we also had stuff on Java, which was
00:39:02not scaling.
00:39:03So, we had to rewrite a lot of that.
00:39:06By the way, Java has now come up with Java 10, where they have virtual threads, which
00:39:12is very similar to Go, but that is still in beta, if I remember right.
00:39:16So, 3 years back, you know, so, those, we had real tough time with threads.
00:39:22But I do not think even if they come up with that, you would not want to shift from.
00:39:26Yeah, right.
00:39:27So, I mean, the problem, you know, with, see, we want to have a small set of very good choices,
00:39:33right, rather than a large buffet of items, right, because if you have a large buffet
00:39:38of choices, it is very difficult to pick the wrong things or a wrong combination of things.
00:39:45In Java, there is, there are at least 10 ways to do the same thing, right?
00:39:50Even writing a while loop, there are 10 ways of doing it, right?
00:39:53Whereas, in Go, there is one way to write a while loop, right, it is called a for loop,
00:39:58there is no while keyword, right?
00:40:00If you want an infinite loop, just, you know, just make the for condition like that.
00:40:04So, it is very easy to, it is very easy for someone new to understand how to do things.
00:40:10It is also very easy for someone to review, because let us say you are doing Java code,
00:40:16you know, you might have to learn about vertex if you do not know what vertex is, right,
00:40:20and then to review, right?
00:40:22Whereas, in Go, because the choices are same, right, it is much easier to review, right?
00:40:27Also, the static analyzers for code are much better now, right, because the code constructs
00:40:35are much limited, right?
00:40:36So, those also give a very good inputs, whether this is going to be a memory leak or things
00:40:41like that.
00:40:42So, we phased out certain things like those.
00:40:47What else?
00:40:48Yeah, we moved a lot to the cloud, right, earlier a lot of the things were in the private
00:40:53data center.
00:40:54We moved almost 80 percent of the workload to the cloud.
00:40:59Yeah.
00:41:00So, how did you pick AWS as compared to Azure?
00:41:05So, I personally worked with Microsoft Azure.
00:41:08Yeah.
00:41:09So, how did you make that decision?
00:41:10So, I am also ex-Microsoft, and by the way, the stuff I used to work on Microsoft was
00:41:15the start of Azure.
00:41:16It was the…
00:41:17Yes, I read about virtualization.
00:41:18Yeah, virtualization, right?
00:41:19Virtual PC and virtual server.
00:41:20So, these were the, some of the core construct on which Azure was built.
00:41:29See, there are a few things, right, I mean, for us, AWS provided the set of constructs
00:41:38we needed to move things fast, because we were running against time, right?
00:41:42We had, you know, the flight was taking off, the business was booming, and we had to move
00:41:48fast.
00:41:49So, AWS, at least for our requirements, was in a better place to help us build that.
00:41:59Let's say, for example, data, we wanted to build a data lake, right, very soon, and house
00:42:05all our data at one place, because there were a lot of analysts and data scientists and
00:42:10analysts coming on board, and they were used to certain things.
00:42:13They wanted clean notebooks, clean data, you know, which is, which never happens, but anyway,
00:42:19so they wanted all these things, right?
00:42:22So, in AWS, a lot of things were pre-baked, where we could quickly go in.
00:42:28And I also, personally, I felt that, at least I had much more experience working with AWS.
00:42:33I knew what works and what doesn't work at AWS, much better than some of the other services,
00:42:42and especially in managed services, right?
00:42:43So, I think the core virtualization infra is pretty straightforward, like compute.
00:42:49What is unique is, for example, managed databases, things like that.
00:42:54Do they give you good insights, alerting, monitoring, right?
00:42:58Because, chalo to easy ho jata hai, right, you start using a database, it's same, Postgres
00:43:04you can use in Azure or here, anywhere, so the code writing is very easy.
00:43:10But can you monitor, let's say, what's happening on your temp DB, to see how your joins are
00:43:14performing, or what are the slow queries, etc., etc., those are the details which matter,
00:43:21for which AWS had slightly better tooling for some of the choices we wanted.
00:43:26So, we did that.
00:43:29And actually, because of that, we moved our private data center also close to AWS.
00:43:33Because what happens is, as trades and trading and prices come in, we ship that to AWS.
00:43:38So, we wanted a pipe, which was a very fast between our private and AWS, and they helped
00:43:44us with that.
00:43:45So, let's suppose there is a software engineer who's currently working in the industry, and
00:43:50he or she is looking towards the CTO position in the future.
00:43:53So, according to your suggestions, what kind of experiences should they get in the industry
00:43:59if they're eyeing that position?
00:44:01First and foremost, you need to be a generalist, right?
00:44:06It's very difficult to get to a leadership position, if you are, let's say, only a backend
00:44:09engineer, or if you only worked on Android, it's very because you would not understand
00:44:15what is happening on the other side, right?
00:44:19And if you don't understand what's happening on the other side, it's very difficult for
00:44:23people to respect your opinions, or, you know, for people to come to you for anything,
00:44:29right?
00:44:30So, the first thing is to be a generalist.
00:44:33And again, that's not a silver bullet for everyone, there are many people who don't
00:44:37want to be a CTO.
00:44:38And that's a great choice, by the way, right?
00:44:40You want to be an engineer, and you want to be an Android engineer, that's a great choice.
00:44:44But your question was, what if they want to be a CTO, right?
00:44:48Then you need to work in different, different parts of the tech landscape.
00:44:54You should spend some time on the front end side, right?
00:44:57You should spend some time on the back end side, should spend some time on the data side,
00:45:02which is again a third big pillar, which is coming up now, right?
00:45:06You may not know everything very well, right?
00:45:09Generally, you have some great strengths, right, either you're a great front end engineer,
00:45:13but you have coded a backend service, and you know what's happening, you know, the core
00:45:18pieces or the vice versa, when you need to have a flavor of all three things.
00:45:24So the second part would be how do you enable people?
00:45:28That is a key thing, because you're not going to code as a CTO, right?
00:45:33And it doesn't matter how well you code as a CTO, right?
00:45:37What really helps is how, how, how can you help people who are actually solving problems,
00:45:43right?
00:45:45This is not like giving design guidance saying, let's do it exactly like this, right?
00:45:51It's all about enabling them to do it better, like a toolbox, right?
00:45:55You don't want to give them a fully baked car, nobody would like that, no self-respecting
00:46:00engineer would take a design from someone else and code it, right?
00:46:06That's not fun.
00:46:07What they look for is a toolbox, like for example, authentication and authorization.
00:46:13If there's a library, great, I can use it, I'll use it how I want it.
00:46:18So enabling those kind of a thing, enabling, you know, a team where people can talk to
00:46:24each other, or an engineer, a front end engineer, a backend engineer, a design person, a product
00:46:30person sitting, you know, pretty much on a table, ideating about the feature, right?
00:46:38People love that, because it becomes a mini company within a company, right?
00:46:42So how can you enable that, right?
00:46:48Stakeholder management is another key part, right?
00:46:51How do you work with your businesses, business partners, product partners, passing on bad
00:46:56and good news as it happens, right?
00:46:59So being able to have those conversations, build relationships, I think that becomes
00:47:06more and more important at the leadership level.
00:47:12So one thing that has, you know, come up a lot in the past few years is AI and machine
00:47:17learning, and we hear about it everywhere.
00:47:20So what do you think the impact of AI and ML will be in the fintech sector?
00:47:24Have you personally seen an example of it impacting the sector?
00:47:28Yeah.
00:47:29So we have, I have, and I also have a lot of interest in this space.
00:47:36So, you know, so when this, so there are a lot of commodity use cases, I think, which
00:47:40are already solved pretty well, right?
00:47:43I mean, customer service, chatbots, KYC image recognition and all that.
00:47:49So I think there are a lot of commodity use cases where there's big relevance and of commodity
00:47:55ML, specifically for in the fintech domain, right?
00:48:03So imagine having a financial advisor, right, who is available 24 7, who knows your portfolio,
00:48:11tracks your portfolio every second, is able to digest every news piece of news that is
00:48:17coming in very quickly and figure out, okay, this news has this impact for Shraddha, right?
00:48:24Is able to understand your preferences, right?
00:48:28And speak a little bit about preferences, right?
00:48:31Every one of us has different attributes, which impact how we make decisions, right?
00:48:38Loss aversion, for example.
00:48:40Let me ask you a question, right?
00:48:41Let's say you, there are two choices, okay, choice A, you get 20,000 rupees, guaranteed.
00:48:51Choice B, it's a 50 coin toss, right?
00:48:55Heads, you get 1 lakh, tails, you get nothing, what would you choose?
00:49:00I'd probably take 20,000.
00:49:01Yes, right?
00:49:02I mean, that's what most people end up doing.
00:49:05But if you look at the expected outcome, right, is 50 50, 50 percent, you know, some 50 percent
00:49:10probability.
00:49:11So expected value from the choice B is actually 50,000, right?
00:49:15And expected value of the first choice is 20,000, right?
00:49:19Now there's no good and bad here, right?
00:49:21It's a choice.
00:49:22And a lot of people have this loss aversion as a attribute.
00:49:27They want to avoid making a loss, even though there's a higher profit at the end of the
00:49:32choice, right?
00:49:34When we give advice, you know, we need to advise it, tailor it to your needs, right?
00:49:44So imagine a financial advisor who knows you more than anyone else, right?
00:49:50Basically what are your preferences in terms of risk taking ability, how much float you
00:49:56need, free cash you need, right?
00:49:59And the same advisor has very good fiduciary qualities.
00:50:03Fiduciary is basically that advisor thinks more for you than for their self, right?
00:50:12You might have seen many people who try to sell you insurance and other things, right?
00:50:16Generally there are some relatives, at least my relatives, there was one of my relatives
00:50:20who used to send me all insurance plans or other things, right?
00:50:23And you know that they have some commission, right?
00:50:27Then you think, okay, if he's selling me this, what commission is he getting?
00:50:31And is he giving me the right choice, are they doing the right choice or not, right?
00:50:36So that fiduciary is always questionable, right?
00:50:39Imagine an assistant who you can trust or whose fiduciary qualities are well defined, right?
00:50:48You would want to have such an assistant, right?
00:50:52And that is actually the promise of many of the things which is coming in Gen AI, right?
00:50:59And we did experiment with some of this, right?
00:51:03And we experimented both with a narrow model, a narrow model is something that we train
00:51:07on specific things, right?
00:51:09I mean, we have a lot of data and let's say company performance, right?
00:51:13If you want to pitch to Shraddha, we'll say we'll look at all IT stocks, there are not
00:51:17many like Infosys, Wipro and all that and we can run some our own ML on them, right?
00:51:25And then figure out what is the price prediction for Wipro or Infosys for that matter, what
00:51:30is the right choice for you and stuff like that, right?
00:51:33And there is a same strategy for many, many use cases.
00:51:37What we found is, but if you use one of the and this is changing very fast.
00:51:42So what I'm saying is for GPT-4, what we found is that actually if you ground it as
00:51:48a whatever I just told you, right?
00:51:50That you are a financial analyst, you have access 24-7 to information, Shraddha's, you
00:51:56know, blah, blah, blah, right?
00:51:59If you train a commodity LLM and ground it to actually, if you sorry, if you ground a
00:52:05commodity LLM into a financial analyst role, they actually do pretty well compared to humans
00:52:13or other things, right?
00:52:14And this is where we want to see how we can augment both.
00:52:18Specific context that we have, context in terms of customers, businesses in India, right?
00:52:29And have a commodity LLM which has, for lack of a better word, real world knowledge, right?
00:52:38So you have, you know, they have in their parametric memory how a financial analyst
00:52:47actually worked for solving, right?
00:52:51And if you don't over define the problem, so we didn't tell GPT-4 to actually, for example,
00:52:58say use sharp, there's a sharp ratio is a common way we measure company, basically a
00:53:05common way we measure opportunity, is this worth investing?
00:53:11If you tell it, then you lose the plot, right?
00:53:15We instead of saying use this, if you say, use a good technique, right?
00:53:24Then we've seen pretty wonderful things, right?
00:53:28So that is something that we, I think is going to change how we, you know, 24-7 financial
00:53:36analyst is something that will be very interesting.
00:53:41So just to get a number from you, how soon do you expect it happening?
00:53:48Yeah, I'll tell you how GPT-3.5 and 4 changed, right?
00:53:52So when we're experimenting with 3.5, you know, let's say we wanted to tell you, okay,
00:53:59Shraddha is X, Y, Z, and a market has fallen.
00:54:04What advice do you give Shraddha?
00:54:06It will tell you a lot of good things, okay?
00:54:08It will say stay calm, don't make impulsive decisions, but it will also give you some
00:54:14bad things or something which could be avoidable.
00:54:17For example, I think sell this, buy this, rebalance and stuff like that, right?
00:54:23It doesn't make sense, right?
00:54:24If you look at GPT-4, the quality of advice is, I mean, the things which are good have
00:54:31stayed, but the deltas have been much, much better.
00:54:34The quality has been much, much better, right?
00:54:38And I think this progression is going to be geometric.
00:54:41It's going to be increasing faster and faster, right?
00:54:45So it's going to happen sooner than most people expect.
00:54:50I think there's a lot of research also people are doing on, I mean, what we talked about
00:54:54fiduciary qualities, right?
00:54:58How altruistic they are, how much hallucination can be reduced.
00:55:02So there's a lot of active research also going on in this space.
00:55:07So I think it's going to be sooner than later, right?
00:55:10The problem with that is if you look at the market, it's an efficient market.
00:55:14What it means is if you have made money, that means someone else has lost money, right?
00:55:21And that's the core hypothesis of a buy and sell transaction, right?
00:55:29And generally what happens is, you know, let's say you find a brilliant trading strategy
00:55:35or a brilliant investment strategy.
00:55:38It's very evident for others, right?
00:55:40And people start copying you, right?
00:55:43And very often that 50% delta, which used to happen with you, right?
00:55:48Starts coming down to the risk free interest rate, which is 7%, right?
00:55:53That's why if you see the top performing mutual funds, you will see some mutual fund X being
00:55:58very good, but only for 2-3 years at max, right?
00:56:03After soon after that, there'll be some new strategy which comes in because the efficient
00:56:07market, right?
00:56:08So if everybody is smart at the same level, right, what would happen is that you would
00:56:15get returns at the risk free interest rate.
00:56:17I mean, at that rate, we'll say which India is growing, which is not a bad thing.
00:56:23So once that happens, then it will be interesting to see how things pan out.
00:56:27Sort of like saying that access to it for all is going to reduce that risk reward factor
00:56:33for people.
00:56:34Yeah, I mean, it's going to save people from making suboptimal decisions, which is good.
00:56:41Say if you are risk averse, right?
00:56:43If something, if your portfolio goes down, it's always reassuring for someone to tell
00:56:50you don't do, don't panic sell, right?
00:56:54Keep investing, right?
00:56:55It's always reassuring for someone to do that.
00:56:59And also if someone understands you a lot, let's say you make bad decisions on Friday
00:57:03afternoons for some reason, right?
00:57:06It's also reassuring for someone, for people to know that there is someone who's looking
00:57:11after me in terms of guardrails.
00:57:14What does it mean in broken world?
00:57:16Let's say I am very impulsive in buying something and I don't know if you know about options,
00:57:22but you can put a stop loss saying I put up, I place a bet, if it doesn't work out and
00:57:27I lose 5% of the money, I want to exit, right?
00:57:31Most people don't put if there's a 5% loss, right?
00:57:34So what happens is they take a naked gamble, right?
00:57:40If there is a friend that I have, a assistant or a co-pilot that I have who knows that and
00:57:46who can help me just before or just after I do something stupid, right?
00:57:52I think that is going to make people lose less money, right?
00:57:58And if you look at the whole investment space, I would say there are two sets of people.
00:58:04There is retail, which is all, pretty much all of us who are investing mutual funds into
00:58:11mutual funds and plowing them through retail and people are buying, selling stock.
00:58:16On the other side are institutions and like hedge funds and all, right?
00:58:21And there is a symmetric situation there.
00:58:26A hedge fund has more skilled people, right?
00:58:29They are maths PhDs and whatnot.
00:58:31They have much better tooling and generally they have much deeper pockets, right?
00:58:36So there's a big skill gap here and when there is this, when there is gap, generally money
00:58:43flows from one side to other more, right?
00:58:48The hope is that some of this tech on this assistance when they come up to speed, they
00:58:57would reduce this asymmetry and make it a more fair level playing field for retail is
00:59:05what I am very hopeful on and working on.
00:59:09Just a follow up question on this, isn't the factor that, you know, fund, people like,
00:59:20institutions like hedge funds or other institutions, since they also have deeper pockets and a
00:59:26lot more access to the newer technology that is coming in, do you think there will be an
00:59:31equality in the future or?
00:59:33So I don't think the equality, perfect equilibrium will happen ever.
00:59:40But with better tooling, the asymmetry would be reduced for sure.
00:59:44We just talked about how, you know, a co-pilot, you know, I think that's a great word, can
00:59:49help me avoid making suboptimal decisions, right?
00:59:54I still use my creativity, right?
00:59:56And I don't think we will have a replacement for human creativity or ingenuity very soon,
01:00:02right?
01:00:03Let's say some stock goes to very high number.
01:00:06This is my belief based upon my analysis and stuff like that.
01:00:11And I want to explore that belief, right?
01:00:14All I need is a co-pilot which will help me do that.
01:00:17If I have 1 lakh rupees, you know, to invest, how much to invest?
01:00:23Do I invest 20,000 or do I invest like all 1 lakh rupees, right?
01:00:29Now there are maths behind this, like, you know, if you are interested, there is something
01:00:34called as Kelly's criteria, which tells you what is the probability of the thing happening,
01:00:39what is the profit out of that thing happening, right?
01:00:42If the probability is low, but profit outcome is better, then you place more bet, right?
01:00:47If higher probability you place, I think you get the gist, right?
01:00:50But most people don't have those formulas and stuff like that, but the hedge fund algos
01:00:56are constantly working on that.
01:00:58So, if retail also has access to this co-pilot, to this kind of constructs, right?
01:01:05Then the skill gap would be reduced, right?
01:01:09And the free money which is, I mean, it's literally free money from many of those folks,
01:01:14right?
01:01:15That flow will be reduced.
01:01:18And I think you must have read in, this is public knowledge, right?
01:01:22made a billion dollars from just the Indian, selling options just in Indian stock actually
01:01:29last year, right?
01:01:30So, those kind of things would stop happening less.
01:01:34So, the promise of AI is less disparity.
01:01:38Less disparity, right?
01:01:40Or more skills for the retail user.
01:01:45Okay.
01:01:46So, this was all about how AI is going to impact maybe the financial industry.
01:01:52Has AI also made an impact on your tech team, on the software engineers working there?
01:01:57Yeah.
01:01:58So, I mean, we use, I mean, we have a good partnership with Microsoft on that.
01:02:02So, we use good GitHub as a core repo for all our code, right?
01:02:07And we use the GitHub co-pilot for a lot of things, primarily automation test cases, right?
01:02:14So, we got pretty good results there.
01:02:18We still don't use co-pilot to auto create code because of some of the performance and
01:02:24security and other reasons.
01:02:27But test cases, we allow it to ride on its own, right?
01:02:31So, use co-pilot a lot.
01:02:34We also use some of the other commodity models which we have on things like predicting traffic,
01:02:41which is very important for us, right?
01:02:44So, because we are a, you know, we are a business at the end of the day, so there are two ways
01:02:49to make profit, right?
01:02:50Make more money or save cost, right?
01:02:53So, when for example, when our infra cost and now let us say this happened on June 3rd
01:03:00for the election day and the votes counting day.
01:03:03Yes.
01:03:04How much traffic would hit us, right?
01:03:07If we have a reasonably good prediction, then we can scale hardware because we are on the
01:03:12cloud, right?
01:03:13So, we can scale hardware for that period of time, pre-scale, right?
01:03:16Auto scale does not work for our use cases because the ramp is too steep, but we can
01:03:21pre-scale.
01:03:22So, if you know the 3rd, 4th is going to be lot of activity, we can pre-scale that.
01:03:28So, we use some of those, I mean we use models of hugging space and all, but then, I mean,
01:03:35you know, so we don't build it ourselves, but we just add, tune it a little bit for
01:03:39our use cases.
01:03:41So, it has changed life for a lot of developers for sure.
01:03:44Ok.
01:03:45And, what advice would you give to a software engineer to, you know, plan out the next 5
01:03:51to 10 years, what technologies should they focus more on?
01:03:55I would say first and foremost stick to the fundamentals and be very good grounded in
01:04:00fundamentals, right?
01:04:02And we just talked about AI, even in AI, I think the fundamentals are very, very less
01:04:10in numbers.
01:04:11They are deep in concepts, but there are very few concepts that you need to know for, you
01:04:18know, for example, you need to know vector calculus.
01:04:22You need to know what gradient descent is, for example, right?
01:04:26Partial differentiation.
01:04:27If you know what those things are at a little bit finer level, right, then you don't need
01:04:32to know there is some Roberta model and for Lange chain, this particular thing to use.
01:04:38That is commodity stuff, you can and that stuff keeps changing, right?
01:04:41And it is not very useful to use that stuff without having a solid understanding of the
01:04:46fundamentals, right?
01:04:47I would say, so, the first thing is stay grounded in fundamentals, very few times the fundamentals
01:04:54change.
01:04:55So, the other part to this would be stay curious and continue learning, right?
01:05:02around, network people, don't stay in your comfort zone and this is very difficult for
01:05:13engineers to do because they don't like to socialize, but I think go meet out people
01:05:19who are in your domain and not in your domain, engineers, let's say, for example, right?
01:05:26So, that to understand how they are solving problems, how things are happening there,
01:05:32right?
01:05:33I think if you are again, if you are early in your career, I would say focus on sharpening
01:05:40your saw rather than cutting wood, right?
01:05:42But there is a lot of time in your career to actually build that, right?
01:05:48I mean, if you sharpen your saw in initial years, then you can monetize that much better
01:05:54later.
01:05:56So, you know, what that means is don't be like, if you get an opportunity, let's say
01:06:00for a slightly better delta in terms of comp and don't just make the decision for that
01:06:05because you might end up in a place where your learning stagnates, you are just doing
01:06:11a point thing in a very big team, right?
01:06:15And at some, and then you plateau, right?
01:06:17So, if you look at it over a longer time frame, that may not be the right decision.
01:06:23One more question.
01:06:24So, you did your master's from abroad.
01:06:27Yeah.
01:06:28Do you think it is still relevant today to do a master's abroad or does it make any
01:06:32change in terms of securing leadership positions, especially for people who are graduating from
01:06:36non-tier 1 colleges?
01:06:39I don't think so.
01:06:40Right?
01:06:41I mean, I graduated about 24 years back and things then were quite different.
01:06:47And I went into, I always wanted to get into academics, that's why I wanted to do master's
01:06:55and then PhD, but life had other plans.
01:06:58I think in today's world, honestly, people look at what you can do rather than where
01:07:04you did your engineering from tier 1 or tier 2 or wherever, right?
01:07:09Honestly, I think those days are over.
01:07:13So, I would say, don't make choices for that because I have this thing on my resume, I
01:07:19will get this.
01:07:20That's very unlikely to happen, right?
01:07:25What is more likely to happen is that if you are able to demonstrate something, then you
01:07:31would have access to a bigger role in the space.
01:07:38One catch, when I was in the startup world about 5-6 years back, yes, there was some
01:07:46bias towards tier 1 colleges, right?
01:07:51In terms of funding and other things.
01:07:53But even that has reduced a lot, I see that now, right?
01:07:58I think people just look at your, I mean, obviously, people put bets on people.
01:08:04If you are an angel investor or a VC, people will bet with people often more times than
01:08:10the idea.
01:08:11With the idea, you will pivot and you bet on people.
01:08:14But I think just making the bet just on academics, I think has reduced a lot.
01:08:21Next question would be regarding the current scenario of tech in India.
01:08:25In fact, this scenario we have seen for a lot of time now.
01:08:30Every year 15 lakh engineers graduate, but not all of them secure a tech job.
01:08:35So do you think the reason is the lack of opportunities available there for software
01:08:40engineers in the industry or is the reason lack of skilled engineers in the market?
01:08:45What do you think the reason is for that difference?
01:08:49So I think there is a mismatch between what skills are needed and what skills are available
01:08:55with engineers, right?
01:08:57As we discussed, the landscape keeps changing.
01:09:01So that is one thing, right?
01:09:03You need to be conversant with the latest set of tools which are needed to craft software,
01:09:13which sometimes is not there, right?
01:09:16So I think there is a, either people are too theoretical, right?
01:09:23And they do not know how to translate theory into code or people are, people have just
01:09:30point skills, right?
01:09:32So they would probably use one specific thing without understanding a little bit of the
01:09:38final aspects of things, right?
01:09:43Data scientists, when I interviewed data scientists, this is one thing that I feel,
01:09:47right?
01:09:48People have just done few projects and they know how, they know what they use within the
01:09:55projects, but not really a little bit more detail on even how the problem was kind of
01:10:01crafted, right?
01:10:02I think many times solving the problems, defining the problem is very important and I see many
01:10:08engineers who are not able to define problems very well.
01:10:13So you believe there are ample opportunities for engineers, just they need to level up
01:10:17their skills?
01:10:18Yeah, I think so.
01:10:20I think there is, even with everything else that is going on and even with lot of automation
01:10:26and other things happening, there is still a lot of scope for skilled engineers.
01:10:35And what is your long-term vision with AngelOne now, especially with the technology part?
01:10:41Yeah, so before technology, right, I think the opportunity in India is immense, right?
01:10:47So if you look at demand penetration, like how many people have actually invested in
01:10:52the stock market, right?
01:10:54It is about 3% right now, right?
01:10:58If you look at China, it is about 15%, look at US, it is more than 50%, right?
01:11:04So the scale is just going to grow, right?
01:11:07You will have more and more people coming in who are going to invest in the stock market
01:11:13in some fashion or another, whether they are SIPs or other things, right?
01:11:19And you are going to have, India has a lot of people who are getting into the working
01:11:25space, right?
01:11:26People are working more, you know, a lot of, India's population is young and actively involved
01:11:31in the workspace.
01:11:33They would have aspirations.
01:11:37So there is a huge space there for understanding customer goals, understanding customers, understanding
01:11:45customer goals and providing them products which are tailored to their goals, right?
01:11:50So the opportunity is immense, right?
01:11:52In terms of the tech space, how do we translate, how do we build products to satisfy those
01:11:59needs?
01:12:00As I said, I think AI and specifically the Gen AI space is going to have a very big influence
01:12:08on this, right?
01:12:10Which would mean that we need to have access to data, that is the first prerequisite, right?
01:12:17Which itself is, we need to be able to preserve privacy for customers, right?
01:12:23So as we get data, we want to make sure that we do not tread on the privacy parts.
01:12:31And with all this, how do we churn out insights fast, because as opportunities are time-boxed.
01:12:41So you know, so those would be the key problems to solve, I mean, and then you go down that
01:12:50route, right?
01:12:51So if you want to solve problems, generally, if you look at context, right?
01:12:54If you share GPT, there is a prompt, right?
01:12:57And the more information you give in the prompt, the better is the response, right?
01:13:02So that is the context, right?
01:13:04So how do we get, and that context space is always limited, right?
01:13:08And actually, you know, so how do you summarize context very crisply for the LLM, right?
01:13:17And how does the, so that it can marry this context with the parametric memory, right,
01:13:24is going to be a good engineering problem to solve.
01:13:30And with initiatives like Gift City coming up for fintech, what changes do you see in
01:13:36the fintech industry in the coming 5 to 10 years?
01:13:41So I do not know enough about Gift City to say how much it will impact.
01:13:49But I would say like again, right, there would be a larger amount of diversity in terms of
01:13:55the products which people would need, right?
01:13:58If you look at mutual funds today, most of them have very high correlation with each
01:14:02other.
01:14:03And even though you feel that there are choices, right, of so many mutual funds and so many
01:14:09mutual fund houses and mutual funds within them, there is a lot of correlation.
01:14:13Most of them would have, let us say, HDFC bank in that, right?
01:14:18And that may not be the best thing for a specific customer's use case, right?
01:14:24You know, for example, if you have NPS, and NPS is already invested for you, you buying,
01:14:30you know, this specific stock, it may not be a best thing.
01:14:35So there would be a huge increase in the diversity of requirements and products as well, right?
01:14:41And this has happened in more mature markets, right?
01:14:44I think people would also look at things like, as people get more and more affluent, they
01:14:50would look at things like capital protection, which is not easy to do, right?
01:14:55You want to beat inflation, and capital protection is not just, if you have 1 lakh rupees, just
01:15:00make sure I do not look it is 1 lakh, right, because there is inflation and things like
01:15:04that.
01:15:05How do you retain capital protection, right?
01:15:08I think those all scenarios will start becoming more and more important, right?
01:15:14And there are a lot of other use cases which are coming up, we are getting, like, for example,
01:15:17personal lending, personal loans, how do we enable customers to get access to loans while
01:15:27managing risks?
01:15:28I think those will be interesting problems to solve.
01:15:32And last question would be, for, you know, engineers who are eyeing the role of a CTO
01:15:39in the future, what is the average compensation, would you say, for a CTO in the industry?
01:15:44I think that is a very difficult question to answer, and it varies a lot.
01:15:52So if, I would suggest, I mean, and that would be, you know, see, I think a lot of the compensation
01:16:01or the rewards that you would get access to is a function of how the business is doing
01:16:09or the product is doing, right?
01:16:11So if there is an advice, I would say, over-index on equity in what you are building for, right,
01:16:20rather than, let us say, a fixed salary or things like that, right?
01:16:25And if you believe in that, and if you have skill in the game, I think compensation would
01:16:31happen on its own.
01:16:33Nice advice.
01:16:36So I think we are done with the questions, and we can wrap up now.
01:16:40Thanks, Radha.