Subscribe
Share
Share
Embed
‘Innovators Playground’ is a podcast series where we explore the cutting-edge innovations and insights shaping the future of technology. It brings conversations with leading founders, industry experts, and thinkers to inspire and inform youth around the world.
DILIP
Hello, all and welcome to Innovators Playground. NPCI’s own podcast series on tech and innovations. We are going to bring the best of the tech minds to you to learn from them. How do you shape your career? How do you plan the future and learn from all these conversations. In this episode of innovators playground, you're going to meet Rahul Chari. I know Rahul for the last 10 years. He's one of the best tech experts who has built the whole PhonePe stack, which is doing billions of APIs on a daily basis. And I'm sure all of you technocrats are going to get great benefit out of this conversation. Hi, Rahul. Thank you so much for giving us your time and welcome to the innovators playground. Rahul, just to kickstart the conversation, how are you feeling now? You know, Flipkart, PhonePe, huge tech build ups, what you have achieved scale, millions of people using it on the road. How has the journey been?
RAHUL
Thank you, Dilip, for having me on the podcast. It's actually great to be here. Every day I am extremely thankful and I feel very lucky, frankly. You and I have had this discussion many, many times where we have gotten to as a payments ecosystem, the role we play as PhonePe, the role NPCI plays as the creator and the operator of UPI, among many, many other things.
It's amazing what we've achieved together in the last 10 odd years, less than 10 years, in fact. To me, this has been the most gratifying part of my career so far. But what excites me genuinely more is the fact that we still have not finished our job. We know that there's another 300 million that we need to actually enable for payments. We know that for the ones who are already so-called digital natives and already are using payments significantly, there's still an opportunity to deepen that relationship much, much more. So, I think for me, the building hasn't stopped. I always call myself a builder. I love building. We've had many discussions on what we can build. So, I think continuing to actually improve on what we've already done, whether it is in terms of scale, whether it is in terms of performance and actually building newer stuff. I mean, infrastructure projects that NPCI is doing so well, building on top of that, creating new experiences for our consumers, creating new experiences for the merchants, continually growing this two-sided network and then delivering on financial inclusion. I think there's a ton of stuff that is ahead of us and that really keeps me most excited.
DILIP
The UPI that was initially designed and UPI now; and I can say that while many things would have been common, but the platform has evolved. When you are building the architecture in the initial days for PhonePe, not only the core platform, but the satellite systems and everything. And what you see it now, how you have actually evolved the platform. And it's like fixing the moving train. And you can't stop the train, but you have to make the train better. You have to make the train faster. So, how has that journey been, especially on the tech side platform architecture and overall ecosystem?
RAHUL
Yeah, no, it's a great question. And to the train example, I keep telling everybody that this is like a Ferrari on a racetrack, but without a pit stop. And how do you actually do any maintenance? One of the things that most people don't realise is everybody is looking at building the next big thing, the next feature, the next platform, the next new shiny object. Incremental improvements at scale, significantly more complex and delivering significantly higher impact.
So, just to all the engineers, especially the young engineers who are always focused on creating the next big thing, I think the challenge of improving on a highly scaled platform for increase in performance is the most difficult thing. And I think that should be a focus for any company like us, any platform play like us.
DILIP
It's like a billion server API happening on a daily basis. ,and you are trying to improve the latency, for example.
RAHUL
Exactly
DILIP
It has a great value for the company.
RAHUL
Huge value. It adds more enterprise value than building anything new. So, I think one of the things that worked well for us is basically our stint at Flipkart. I have to mention that when we worked at Flipkart, and it's not just Samir and me, I think a lot of my engineering team, at least the folks who joined me very early on, had some sort of stint at Flipkart, as well as other companies that have built platforms of scale in India. So a big billion day at Flipkart, huge learning in 2014 when we launched. Similarly, a bunch of other things, flash sales that we launched for mobile phones.
DILIP
And the volume certainly goes 10x 20x.
RAHUL
It's the herding effect where both concurrency as well as number of transactional calls that happen in a single second absolutely shoot up.
DILIP
I can't resist myself from asking this question. Every transaction ended with payments. And UPI wasn't there. So, how did you guys manage payments?
RAHUL
We didn't manage payments. In fact, I don't know whether you know in the first 15 or 20 minutes of that event, pretty much every payment gateway in the company actually came down in 2014. And that was one of the very important reasons why we started PhonePe. Outside of the fact that COD at scale, managing cash at scale is a problem.
Payment gateways were not holding up. So, infrastructure payments were not solved for the digital transaction load. That was one of the reasons we started PhonePe. But coming back, all the learnings that we had from Flipkart and building platforms of scale, we actually applied at PhonePe on day one. Some of the things that I think held us in good stead were very key decisions that we actually did. We have something called ‘shared nothing architecture’, wherein all of our databases are basically isolated to only the services that actually operate them. And any sort of insert, update, read for those databases actually happen only from the services that own them. .
DILIP
So, kind of create a layer.
RAHUL
You kind of create an application layer over every database and make sure the databases don't become abused by carrying unnecessary data just to actually optimise your performance on the edge. It gives kind of returns in the near term, but in the long term, it becomes absolutely impossible to manage. What that means is that you have to actually have significantly more network calls to actually solve for possibly one single experience on the app side of it. But that is actually better than creating what I call as the postman effect, where you collect a lot of data into a single database and the database becomes the choking point. Ultimately, you live and die by the scalability of your database when it comes to transactional systems. Decisions like that we took very, very early on. Some of the other decisions that we took on the database were to be sharded from day one, because the biggest nightmare you can get into is to completely re-shard your system. Early on, it might seem unnecessary. But some of these basic common sense calls when you're actually building out the platform, if you're building something that you're going to be delivering population level scale becomes extremely important. The other things that we did were to actually build what I call as UX driven services, where all your API and the API models are designed from a consumer flow lens, not bottoms up from what the service wants to deliver. Because if you do that, then you actually become more optimal in terms of the number of service calls that are ultimately making for any experience on the app. Whereas if you actually built up saying that, okay, I have this service, this service can deliver these different primitives, and I build an API for every primitive, somebody is playing an orchestration and a composition role on top of that, which adds unnecessary performance.
DILIP
Very similar thinking, you know, in fact, when we were looking at UPI and the market was asking that, while yeah, this is the way we connect back to us. Whether it's UPI or any other payment system later on, which migrated back to the unified specs, what is the next mile or for example, the app to the central system API, I used just give one answer, just relay the UPI specs, correct, right? It becomes very simpler for the ecosystem. And you're talking about the same thing.
RAHUL
In fact, it was love at first sight. When you looked at the UPI specs, the extensibility of the UPI specs was so beautiful. With that simple structure of separating the authentication and authorisation model. We in fact, when we started PhonePe, Dilip you would know, we actually started building something like UPI on top of IMPS, because IMPS didn't have the pull and the push. Very similar to what chillr at that time.
DILIP
I remember that.
RAHUL
And then through Nandan and Pramod, we stumbled upon UPI.
DILIP
Yeah, if you recollect, we met at Taj Colaba. at the UPI launch.
RAHUL
I remember, yeah. Also at the hackathon that UPI had actually done very early on to encourage the ecosystem to think about what all they can build. So yeah, those are the kind of simple principles that we've actually taken forward. It doesn't mean that we don't re-architect or refactor. But what it means is that we have managed to actually stay ahead of the business as far as possible when it comes to scale, that you have enough time to actually refactor the code versus get into technical debt. Even today, we do refactor.
DILIP
Also from app standpoint, was the app designed like a dynamic app right from the beginning, or you kind of moved over there. You opening up the app, you get the downloads, upgrades almost like every time you open the apps, that kind of gives you a great agility or advantage in terms of time to add new functionality or keep making the app better, as in when consumer is opening it. So, it's a delicate balance, right? I'm just saying how much you…
RAHUL
Yeah, you're right. In terms of how much you can actually push to the app dynamically versus how snappy the app is when it actually opens immediately. One of the principles that we also had in building out the app that the experience has to be extremely snappy. We have a cut and forward model, which means that everything that you see on the app will only happen if the app has already synced and pretty much downloaded it. It doesn't actually make it, most of the stuff that you see on the PhonePe app is not necessarily rendered by just calling the server and immediately rendering that content. The app itself is a fairly intelligent ecosystem by itself. We have a complete sync server inside the app that has intelligent prioritisation of what is the launch screen that needs to get synced in line, and what can be lazy synced. We have in the network layer, we have priority queues to say that which are the calls that go into priority zero, which are the calls they want to priority one, which are the calls that go into priority two. We have a database management layer with its indexing on top of it so that you can actually build search there. So, it's important to think of the app also intelligently as a full fledged ecosystem, and not just the view layer, because that's what gives you the flexibility to innovate and keep the app snappy.
DILIP
I think this whole client server interaction, you know, I think you have taken it to the next level. In that sense to get the client also have the intelligence at the same time, continue to work as a client. Because tomorrow, if you create that download layer intelligent technically, the client can operate as a client and client can also operate as a server.
RAHUL
Either way, if you look at it, the compute power that all the phones have is only going up. It's only going up, right? The cost factor is going down, the compute power is going up, which means that if you're not utilising that to the max, it's a criminal waste. There is an edge server, if you may call it that is available, how best to use it because that can deliver not just performance, it can deliver privacy, it can develop much better controls to the user. You can actually give guarantees to the user in many ways to say that there's certain amount of data that is not leaving your ecosystem. So, there are big advantages of looking at that device as a server and how do you maximise it. Your consent architecture can have some parts of it on the edge, and only some parts of it on the server side. With AI proceeding the way it is proceeding, one of the most exciting things for me is to look at whether there are tiny LLMs that can actually start getting hosted on the phone itself, while the sizes are still pretty large. But there are tiny llama models that are already coming about that is sub one GB for 1 billion parameters, while it may not be ideal for doing general queries, but you actually tune it for domain specific queries, you might be able to actually host this on the device itself. Now, it still serves only a percentage of the devices that are out there. But it's only a matter of time. And at the rate things are progressing very soon, that some of these will be hosted on the edge. But the larger picture being that this particular device in our pocket is getting more and more powerful and how do you utilise it.
DILIP
While we have been talking about the app, the UI, the whole consumer experience, how do you think this generative AI is going to kill user interface? What do you think about it? All the colorful apps are going to become like a vanilla ice cream or like a one prompt and you know, people will just talk to the app.
RAHUL
I think 100% the way we interact with technology will change, it would be wrong to actually say that this is where the way things are going to be working forever. to be very frank. last year to this year, I've also started significantly shifting towards saying that yes, AI is very, very real, it is actually going to disrupt, and there's going to be a fair amount of change. Now the question remains though is even in the past, there have been some of the modes in which you actually interact with devices being voice, Alexa was significantly voice enabled, of course, the intelligence to actually orchestrate a transaction or an experience was not perfect, while the understanding was still good enough. In some cases, the understanding obviously was very poor. Now the general understanding by the models of voice driven commands is gone to a new level, different level. Now, agentic orchestration of the same using some of these models across a transaction, etc, also is getting there, I would still believe that there'll be parts of the experience that continues to be the way it is, because users do like to actually be presented with certain amount of choices, make a decision based on those choices. Some of which is highly transactional might actually change. In which case, yes, we may actually bypass the interface layer as we know it today and go directly to the back end to actually just execute the transaction. But where choice still matters significantly, and you're not ready to give up the choice, I think you will continue to be with this kind of experience or more evolved experience.
DILIP
I just hope you know, the colorful apps remain as they are today, I think. And that's what is the journey to the super apps.
RAHUL
When I talk about choices, if you look at menu driven operations, now, if you have to present every choice, even whether it's voice based, etc, it is you'll never be able to process it as fast as the human brain by just seeing it can do it, right. So that's why I believe that it's going to be a mix. It's always going to be a mix, in my view.
DILIP
You know, China has progressed in ways very different way than what India is. And, you guys have started the journey and I believe this is still work in progress on super app. So, what's your thinking?
RAHUL
I call myself now a company of super apps, not a super app. We have PhonePe, we have Pincode for hyper local shopping, we have Share.Market for trading, and we have Indus app store.
DILIP
So India is not a super app country because of the regulations, laws or because you think it's it's fairly safe to operate that way because you know, everything is innovating so fast.
RAHUL
I don't think it's any of those. I think India is not a country of super apps because users in India are very different and they like to actually compartmentalise their experiences to separate apps. China, I believe is a very different user base. You can in a single app throw everything at them. You can have flashing banners, you can have different offers, you can change the layout. Yeah, they are more used to that and like that. I don't think the Indian user base is like that. Right? It could be for various reasons. It could be the way our journey into internet ecosystem has happened. It is still happening. People are coming in it through content and content consumption. And content consumption is compartmentalised to certain apps, then they get exposed to shopping through another app, etc. Whereas in China, the journey possibly was all through a single app right from very early days.
DILIP
Yeah, I think that was their starting point in that sense.
RAHUL
Correct. How this journey has happened and the fact that we do like to compartmentalise will mean that I don't think there's going to be a single super app that takes the entire spectrum of content, shopping, payments, financial services, etc. There will be a certain amount of aggregation, like a PhonePe is a payments and financial services app now. But shopping, putting shopping into PhonePe, for example, we pride ourselves on PhonePe as being our session sizes are sub 15 seconds, we have to be as good as that if you're going to be the best UPI app out there or the best payments app out there. Now to force a shopping experience on top of that, which is largely browse and experience driven is a collision of core goals on the app itself. My view personally is that there's not going to be a single super app in India.
DILIP
Rahul news on DeepSeek building the LLM model in five million dollars. How do you see the impact of AI on the overall ecosystem and PhonePe? How much do you use AI tech people have started using for coding? How does this may shift the landscape of what we are looking at?
RAHUL
Last year, I still remember in one of my tech all hands, I had said that code generation, yeah, using any of the AI tools is a big no no. Yeah, yeah, I said that last year, early last year,
DILIP
Are you still on it?
RAHUL
I'm still officially on it. But I'm not on it per se. Because I was genuinely worried about the fact that if we actually generate code that is going into production, especially in the environment and the regulated environment that we operate in, two concerns, one explainability of the code is something that I want to actually own and I don't want to give up on. And second one is in the case of any sort of incident, does it actually put us on a handicap. Because you don't know enough about the code that was the concern.
DILIP
Your fear is the developers may go blind with this.
RAHUL
To some extent. But over the last year reading a lot seeing how far the ecosystem has come in terms of the abilities of the AI tools from a development perspective, I think I have to relook and we are actively relooking every part of the SDLC process.
DILIP
Some other tech founders are saying that they have deployed it and the extensive amount of code is now written by AI
RAHUL
I think 100%. If you look at functional snippets of code, it is extremely good. If you look at a large-scale distributed system where you have to actually have it well tuned by interaction across various functional codes, it is still a while to get there.
DILIP
Interdependent issues will still have to be dealt by the people.
RAHUL
Large scale distributed systems have to be still designed by software engineers. Whereas code snippets, large functional blocks of code can be actually assisted and generated. We are looking at both coding assistance for the IDs that we use. Copilot is one of the obvious candidates, we are looking at pull request generation through one of the tools because many times there are pieces of code that you need to get integrated across the entire ecosystem. Each of those integrations are very similar with only minor nuances by the client, you can actually do that seamlessly. Unit test generation is something that engineers are already using it across the board very well. So, mapping the SDLC process into aspects that can be leveraged by these tools and then looking at which are the tools that we can actually authorise in the company. That's where we are at. The other thing that we are planning to do is outside of just authorising which tool, any access to the LLMs or even something like say Chat GPT for snippet code generation or for optimizations, go through a platform layer that we built where there is authorisation and audit trails as simple as that. And then over time we can actually even block certain types of data that are getting submitted, even though this is under the contractual obligation of say Azure OpenAI where they're not going to be using it for the training of the model itself. That's where we are at as PhonePe. But I must say that it's come a long way in a very short period of time.
DILIP
And it actually scares me the way it’s progressing.
RAHUL
And I think it is while it is actually extremely useful, productivity gains have to be measured at least at PhonePe. We are not yet there to actually claim that this is having a significant impact, but maybe by the end of the year, we will have a very strong picture on that one.
DILIP
No, no, I think very rightly said we at NPCI also are looking at a, you know, if I may call it as a calibrated approach towards this, at least I have started feeling there is no escape from this.
RAHUL
No, there is no going back. And I think software engineering overall will change. In the long term, how this shapes up and what it means for software engineering as a profession, I think only time will tell.
DILIP
I think this whole concept of full stack developer will gain the importance because if you are saying that as a modular or a very functional code, if the AI can write, then technically that can be the developer's job in that area can be replaced by the by the tool and the developer can actually step up and actually look at a much larger area under a zone where is integrating with multiple components and full engineering view. I think that's where I think the engineers should move.
RAHUL
In fact, I think more and more of software engineering outside of deep research has become a generalist role. I think the way engineers should look at it is to get more and more into specialised role.
DILIP
Yeah. Now, how do you handle such a large, talented workforce? I've only read this, you know, Bangalore has a has a bug off this work from home. What is your view on this?
RAHUL
My view actually is that work from home for engineers is not a good option, right, just from their career perspective, as well as from a PhonePe perspective. In fact, as PhonePe back in 2021 itself, towards the end of 2021, we had a back to home city. Then in 2022, we got into hybrid model, where it was two or three days in office, and then two or three days at home late in 2022 itself, we started actually saying that it's back to work as always. Work from home was a privilege even before COVID. Like many companies, we had a work from home option where you could exercise that privilege for a variety of reasons, sometimes on the personal front, and sometimes because there was either a migration that you are working late into the night, but a planned work from home because I want to do it. I don't think it's a good option for PhonePe. The life stage we are in, we are still a growing company.
DILIP
But I think, you know, very right. I'm saying if you need it, take it right if you if you are personal front or whatever is forcing you to do that for a short period of time, please go ahead and take it. Yes, because that's a more like an employee empowerment, you know, the employee care for that matter.
RAHUL
Correct. However, in general, I feel for young engineers, especially your learning curve and your growth curve significantly reduce because you become like I said, you then become a developer. If you're a software engineer, you're working sitting next to the business guys, you're learning with the product teams with the analysts, having those conversations, problem solving end to end. Otherwise, you're a developer. And again, coming back to the AI revolution, if you're not a full fledged software engineer collaborating with people, you're getting commoditised as a developer through AI…
DILIP
And which AI will replace one day. No, I actually very rightly said and you know, all the the engineers watching this, I think don't look at a short term gain, look at a long term view. I, I think this whole AI thing coming up, you know, I think the the need of upping your skill upping your interactions and, you know, learning through various mechanisms is so high. So, it's better to be in the office than to getting commoditised and become multifunctional.
RAHUL
Yeah, yeah. Selfishly, though, with Bangalore traffic, I hope that there are other companies which are allowing work from home and PhonePe engineers can come to office. That way it's a win win.
DILIP
Rahul, we I just learned that you and my brother were from the same batch. I heard that you got into IIT and it didn't take that call, very unusual decision. I may say so. Right. The IIT brand is still the IIT brand.
RAHUL
That's because I was not good enough to get a rank that would get me into computer science or computer engineering in IIT. This has become one of those clickbait headlines. Unfortunately, from one interview where they just asked me that a lot of founders are from IIT. Did you go to it? And I said, No, I went to Mumbai University, Sardar Patel College of Engineering, which I love. And they asked me why. And I said that I did give the exam. I didn't get a rank. I was getting civil engineering in Bombay, IIT. But I was hell bent on doing computers. I chose it. And that was the answer. But that's still the simple answer. Nothing else. My parents were very supportive. They never had a problem. But interestingly, when I joined Sardar Patel College of Engineering, a lot of students were shocked that I didn't take IIT. And I still remember one of the professors who himself was from IIT said that I had made a terrible mistake.
DILIP
Really?
RAHUL
I remember in my first year, then I ended up writing an article in the college magazine about why IIT is overrated. That was my one rebellious moment.
DILIP
I think it's good to have non-IIT founders and I'm somewhere trying to brag myself as well. You know, while there is so much tension at work, so much pressure at work, and this pressure is we have put it on ourselves. That's the way I look at it. You know, because there's so much to be done. How do you wind down? How do you spend time on the weekends? How is the work life balance?
RAHUL
It's only recently my Flipkart stint, and a large part of my PhonePe stint, I had not really had a great work life balance. But at PhonePe now we have an excellent team, very proud of the team. And I think everything aside, tech, decisions to go all in on UPI and everything else, the only thing that sets us apart continues to be our phenomenal team. And I'm very proud of that. They've given me the luxury to try to take a more balanced approach to work at home. My hobby actually is very interesting. It is nothing to do with tech or digital. I love carpentry. I'm a woodworker.
DILIP
There also you’re a builder.
RAHUL
I'm a builder. And I love the fact that it's a precision hobby. It's a lot to do with drawings, exact measurements, etc. I build, I even build furniture.
DILIP
No, no, very good. I think, you know, as you said that, you know, if you if you love, you know what you do, normally the balance automatically is derived, right. Compared to, you know, if you're forced to do a work. And that's what I also believe. I see few other the tech expert like Pramod, you. And I feel that, you know, all of you guys are like, like the core pillars of, you know, the digital India in our prime minister's vision. How the younger engineers or new engineers can learn from your benefit? Why don't you start a tech blog? Why didn't you write more or tweet more? Or why don't you give benefit of your experience to the younger generation?
RAHUL
One of the person who has been pushing me the most to try and open source has been you right. And we've had this discussion many times. And NPCI has already started doing that.
DILIP
We have started the journey.
RAHUL
You started the journey.
DILIP
Because with the simple thinking that, you know, we have been a biggest beneficiary of the open source, right? I think we go back to the ecosystem and we need to give back something. I think that's the thinking.
RAHUL
No, so we've also started that very recently. In fact, we've actually moved away from MI source as our container, orchestration and resource management layer and have an in house platform called Drove, which we've just recently open sourced. So, we have started our journey on more and more open sourcing, we have actually started having a significantly rich tech blog about everything that we are doing. In fact, it has pretty high level of viewership. The feedback there is to actually get even more in depth about the tech versus being more product and tech as it is today. So, we have actually taken that feedback and we're going to actually go even more in depth on how we build the tech. So, I think that is one of the things that we are taking very seriously. I do believe you're totally right. I think if we don't do that kind of give back after having been a recipient of open source for so many years, I think it's just wrong. From all the engineers, one of the things that I keep telling everybody is that especially now with AI and the barrier to actually solve problems using technology coming down even more, right? There was a whole phase of SaaS and public cloud, etc, which brought down the barrier to entry to build technology. Now that has been crashed further with the potential to actually use AI to basically generate code, right? The number of problems around us as a country and the fact that people like you, Pramod have proven that there is a API driven model that can solve problems at population level scale. I think just looking around that, finding what is the problem that excites you because tech is no longer a barrier. I think that's the way for every engineer to actually look at this.
DILIP
Being said that Rahul, I would still urge you to be available as much as possible, not only to the PhonePe engineers, but to engineers of the country because you know, you're so brilliant. And I believe that you can create a lot more value to India's aspiration of building tech for the world. Rahul, I can't let go without the rapid fire, which is going to be a fun round. You can't think you have to just answer from your muscle memory fast enough. What's your favorite job till date?
RAHUL
What I'm doing right now with PhonePe.
DILIP
Who's your role model?
RAHUL
You, Pramod, Nandan.
DILIP
Reading or podcast?
RAHUL
Podcast.
DILIP
Movie in a hall or…
RAHUL
In a hall. Always.
DILIP
The latest Gen Z terminology you learned and this is a very controversial subject by the way.
RAHUL
No, no. My daughter is 15. The latest one I learned because she is giving her exam is cook or be cooked.
DILIP
What do you think the next innovation is going to be?
RAHUL
I think we are living it right now. It's basically the impact of AI across the board.
DILIP
Thank you so much Rahul for, you know, spending time with us. I just believe that, you know, we have so much to do assuming that the PhonePe and PCI and the entire ecosystem will continue to make India proud. Thank you so much.
RAHUL
Thank you so much. Thank you.
DILIP
Thank you all for listening to us and you know, I hope you like the conversations. And do let us know your feedback in the comments section. We will try to come back to you as much as possible.