Dhruv Sharma: Hello, and welcome. This is stream 140. We're live. We hope you're live. And if you are, then, you know, come in the chat, hang out with us, bring up questions, and we'll be sure to ask our guests. We're covering three, different sectors today. We're doing voice first with Raoul of Navana.ai, and then later we're doing payments and health care as well. Raoul, welcome to The Offline Network, stream 140.

Raoul Nanavati (Navana.ai): Thanks. Thanks a lot for having me. Thanks, Dhruv. Thanks for Utsav.

Dhruv Sharma: Fantastic. Introduce the company, Raoul, and then we'll take it from there.

Raoul Nanavati (Navana.ai): Yeah. Sure. Navana.ai is a sovereign voice AI company. We focus on, building voice AI infrastructure to connect businesses and governments with, end citizens. Our main focus is on working with regulated industries today, and, building across all Indian languages. So our main focus is on solving for Indian language complexity. So we actually build from the model up. So we've created models for speech recognition, being able to recognize the way we speak in India. So that's multiple languages in a single sentence, background noise, multiple speakers—being able to tease all of that out and recognize the way we actually speak here, and then also to be able to synthesize text into audio, in all these Indian languages. Right? So this is where the big tech players have not spent a lot of time, solving for this. We've spent a a number of years actually solving for this complexity, and now we've verticalized this into a AI call center product where we're able to, you know, deliver the same level of quality, whether you're, you know, Raoul sitting in Bombay speaking in English or someone sitting in a village speaking Magahi or Bhojpuri, Odia. Doesn't really matter. You get the same level of service and access, as a result of the product that we deliver.

Utsav Somani: And you're doing this with your brother. ₹40 crore in the bank, so congrats on that, led by Ronnie Screwvala. And

Dhruv Sharma: Thank you.

Utsav Somani: Eight years. I mean, so you've not been doing AI just for those eight years. Right? I mean, you must be doing something else.

Raoul Nanavati (Navana.ai): No. So, see, the mission of the company has been to make, digital services accessible to the next billion user market. This was the first sentence. This was the first thing we ever created at the company, this one sentence. This was when Jai and I, we both did our MBA from Cornell Tech together. So we were in in New York thinking about this major problem, in India, and, came back once we graduated, to to solve for this problem. And actually, recognized back then, in 2018–19, we had a phenomenal design partner in Ujjivan Small Finance Bank, to solve for their problem of getting their 5,000,000 low-literate female customers all across the country online. Right? So we really looked at it more as a design problem, then spent about nine months working on this specific problem. And what emerged was voice as an interface back then, 2019. And then, you know, we happily went out to Google and Microsoft to find APIs to build this product, and those APIs didn't exist. And so then we had to sit down and say, okay. Like, are we now actually gonna build are we gonna become an AI company? Are we actually gonna build these models? Or are we just gonna wait until the technology matures? So we chose at that point of time to actually start working on voice as a core problem and, actually moved into becoming a core AI research lab. So that's what we did for the first four years. We collected data across all, you know, the languages of the country, created models, and, you know, spent a good chunk of time just doing that core research work.

Dhruv Sharma: Tell us what that journey was like, Raoul. Like, how did we go finding training data to, you know, basically train your own proprietary models?

Raoul Nanavati (Navana.ai): Yeah. We were, very lucky to work with a number of fantastic organizations. Right? So our first design partner, back in 2019 was, a small finance bank called Ujjivan, led by a chairman called, Samit Ghosh. So he was a visionary. He he understood what we were saying and the vision that we had. He was as audacious as us and said, you know, like, if this is what you're looking to do, we I will help you. And so our first, you know, network of of data collection, actually took place with with the Ujjivan Small Finance Bank staff members. Right? So and they had staff members all across the country. So we created the entire network. We worked with them to collect, you know, the first seed dataset. That served as a proof of concept, which we then replicated with Microsoft Research, with IISc Bangalore, with IIT Bombay, spent, like, a number of years, working on then using that kind of infrastructure that we built then, to create an open-source data. So, like, even with IISc Bangalore, it gets a foundation funded project with them. We released 15,000 hours of data across 45 dialects using the same infrastructure. Right? So alongside building a number and releasing a number of scientific papers solving for a lot of the issues, that are unique to India. Right? We were talking before we we came got live, that even defining the number of dialects in India is like the is a problem. Right? So if you're a scientist looking to solve a problem that is not even identified in terms of categories, you know, you're at a really, unique place, to solve for some of these problems. So, yeah, I think it took a a a good chunk of time, effort, perseverance, and then finding what kept us going was that we wanted to find the most impactful channel, where we can actually create this combination of voice interfaces to quality of life services. Would that be WhatsApp? Would that be, you know, voice notes on WhatsApp? Would it be calling? Would it be a bot on a website? Would it be an operating system? We went through all of these different kind of channels over the years, figuring out what is the right, you know, most impactful channel. And, you know, AI call centers ended up being, like, telephony. Actually, just picking up a call, making a missed call, getting a callback, and getting access, and now being able to speak in your language, whether you're illiterate, low-literate, literate, whatever language you speak, you can now get access, do a transaction, apply for, you know, a government scheme. All of this now through a telephone call at scale, no matter how many people we have who are calling in at any time of day. I mean, it's incredible, incredible moment of time that we're in, the power of the technology in terms of actual impact on ground. We're about to see something amazing happen.

Utsav Somani: I'm gonna sidetrack a bit because you mentioned voice notes. Dude, there is a class of people I just discovered on Twitter who basically hate voice notes.

Raoul Nanavati (Navana.ai): Yeah. I'm like, man, like,

Utsav Somani: you can control, you can transcribe, you can do it, you can 3x and listen to them. But have you come across such people?

Raoul Nanavati (Navana.ai): Oh, man. My my brother is gonna kill me for saying this, publicly. But but Jai, my brother, is one of those people. Finally finally enough, I mean, we were on ground. Right? We were doing research for the first, eighteen months of the the country. We touched, like, you know, 10 states and, had, like, one-and-a-half-hour long interviews with, like, 500 people, right, to understand what is this emergent smartphone behavior that people have. Right? Like, this is the first few years where people had smartphone, first few years where Internet was free. So it's like this deluge of people who had, like, this crazy access to all sorts of stuff, but no ability to understand the interface. Right? Yeah. Because all in English, all Western made. So we actually went to see, like, what what is working. Like, what out there is already working. And one thing was voice notes on WhatsApp. Like, you'd see people on WhatsApp, like, not a text inside voice note, voice note, voice note, voice note, voice note, voice note, back and forth. Right? But, yeah. This class of people really don't, appreciate the fact that, they, you know, the other technically

Utsav Somani: super efficient. Like, I think

Raoul Nanavati (Navana.ai): they can just get the messages off,

Utsav Somani: like, using Whisperflow and 10,000 different tools to communicate I'll

Dhruv Sharma: tell you what people don't like them. It's because for the sender, they can speak faster. They can write, but the receiver can read faster than they listen.

Utsav Somani: But now there is transcribe feature. There is 2x voice note or 3x voice note or 1.5x of voice note. So, I mean, there are hacks around it, but I know the effort shift then to the listener. Yeah.

Raoul Nanavati (Navana.ai): That's that's exactly why Jay doesn't like it.

Utsav Somani: Yeah. Tell us about the scale of the problem, though. I mean, how many dialects are there? We were discussing this before the show. There are so many different languages. You're covering 15 of them right now, but it's a massive how do you get training data for this? How does it become better, your models?

Raoul Nanavati (Navana.ai): Yeah. There's no easy answer, to solving this problem. The reason is that the rest of the world had been seeding and generating data as a result of using the Internet for the last 30 years in their language, German, French, Spanish, English. I mean, you have millions of people for 30 years just as a function of of using the Internet generating this data. We have not been doing that for the last 30 years. Right? And so just, you know, taking a trawl and, like, sweeping out all the data and scraping it off the Internet and then putting as much GPUs as you want behind it and then, boom, you have a model. That option just doesn't even exist today. And so you actually have to create the data. So we have actually gone out and created the data. So we actually reached real people. We actually identified, you know, real domains that need the data. We went to domain experts. So, like, in banking, a domain expert was like a bank branch manager in a village. We go to them and say, hey. You know, for, you know, bank balance questions. Like, give me 10 ways that your customers ask you in colloquial conversational language about their bank balance. You know, loan Right? Like, there are different ten ten ways, and we got thousands of these domain experts to generate these sentences. Then we fit those into an app, then we find people in the same locality to speak out those sentences, and then you have data. Okay. So this is what we're talking about in terms of actually now building and solving for this real problem today as opposed to just scraping the Internet and then and then going. So this is why there's this gap and why companies like mine can exist when someone like Google or someone else, is out there. Right?

Dhruv Sharma: And isn't that ironic for a civilization that had an oral tradition where all of our civilization of history is also not documented. You you've had a tough time going to the computer. But tell us this, Raoul. I would assume that by now, the models have gotten pretty good, but, over time, but there are there are still use cases where they fall short. Let's spend some time talking about that as well.

Raoul Nanavati (Navana.ai): Look, I don't think it solves them. In English and Hindi, I think we're almost there, right, at a global level. But if you go beyond Hindi, it's still not a solved problem for, like, many things. As a company, Navana, we we have a competitive advantage because we are actually, like, functional in South Indian languages versus our competitors because they depend on on global models. Right? And that has come primarily because of the data that we've we've collected. Assuming, you know, moving beyond that as well, if you were to look at just domains where there is specific terminology, health care, for example, you're mixing in now health care, you know, terminology along with mixed input of English and Hindi or English and Tamil. Like, it's a whole different language completely. Like, we have a banking specific model that we have created for speech recognition that now understands interest rates, numbers in all of these languages, and in the way that we speak it. We'll speak all the way in Tamil, and then when we speak the number, we'll speak in English. Right? So, like, all of these idiosyncrasies, that, you know, is beautiful when it comes to, you know, India and our our, like, complexity and richness, those are the things that need to be accounted for, in the models. And then, of course, things like health care terminology, banking, numbers, numeracy, these types of things.

Utsav Somani: And you're working with banks. I mean, so we've spoken to a bunch of voice AI startups and your space card, like, rightly, Dhruv said as well. All of them bring BBBFSI as a sector. I'm guessing because of the being in the revenue pool itself. And you would on-prem, which I think might be a differentiator also.

Raoul Nanavati (Navana.ai): Yeah. Yeah. Yeah.

Utsav Somani: To a company more model. So tell us a little bit more about why why APIs versus on-prem.

Raoul Nanavati (Navana.ai): Yeah. See, the whole last twelve months was all about does voice AI work. The answer is a resounding yes. And now that it is a resounding yes, now people like the CISOs are getting the, you know, the people who are in charge of security or the people, you know, the information security folks are now getting involved because now the mandate is no longer experimentation. It is now pan-India deployment at scale. When you start talking about that, then privacy, DPDP Act, requirements, all of these things start coming to the conversation. And the simplest solution to being completely compliant, at least on data privacy, and security, is to sit within the infrastructure of the customer. Right? Then you're you're not going in and out. The data is not moving anywhere, and you're totally compliant. So that on-prem aspect, I think, is something we're gonna see, as a need moving forward. And because we own the models orchestration layer and the agents, we just pick it up, put it inside the bank infrastructure, and we get moving.

Dhruv Sharma: Let's go maybe one level deeper, within the context of BBBFSI. It'll help us understand the unit economics. Help us understand how those voice minutes really translate into economic ROI for the institutions.

Raoul Nanavati (Navana.ai): Yeah. For sure. What's up your question? We want in BBBFSI. I think revenue pool is one, but if you were to think about the calling of the country, First, regulated industries account for 80% of all calling between an entity and a human being. Right? So whether it is the government, BBBFSI organizations, health care companies, this accounts for 80%. Within that 80%, 80% is coming from BBBFSI. Right? And that's why everyone is there. Right? All of the call minutes that really matter today and will matter for the long term sit there. What does all of this calling that is happening, Dhruv, you asked. Right? So there's four real thing the four main things that are happening. There's sales calling. Right? So Okay.

Dhruv Sharma: Block for

Utsav Somani: Bajaj Finance. Like, I think that's the question that you should answer for everyone. How can we block calls from Bajaj Finance?

Raoul Nanavati (Navana.ai): Yeah. They're one of our one of our customers, so my apologies. Shared with shared with shared, and

Utsav Somani: the product is working too well then?

Raoul Nanavati (Navana.ai): Yeah. Yeah. So sales is a big one. Right? Sales collections is another one. Right? So collecting, for that compliance calling. Like, for example, once you disburse a loan, you need to get confirmation from the customer that they have received that exact amount, and you need it on record from a compliance perspective. The number of calls that need to happen in that manner and there's some compliance plus, kind of calling, like, you know, setting up or refreshing your KYC. Right? These types of things. And then the fourth is customer support. You're calling in. You you need some support. So these are the four main areas that Navana works in. It is the the four main areas that voice AI is playing in today, at least in the BBFSI.

Utsav Somani: And about the unit economics?

Raoul Nanavati (Navana.ai): Yeah. So, I mean, if you look at the unit economics, it kind of speaks for itself. Right? A human being cost anywhere between ₹12–15 per minute just basis the wage, not assuming any other cost related to handling a call center. And, voice AI is coming in at in the last three months has been, you know, a major price for it's now at about, you know, ₹2.5–3 a minute. So just between the difference of cost there, as we move forward, the quality is already at, you know, at at the top of your human agent performance. Right? So maybe not like the number one, but in the top band, probably the lower end. Right? Like, maybe an 8/10. And it's consistent 8/10 today for all of these use cases. As we move forward, it's just gonna get better and better to very soon, it's gonna outperform the top agent top human agent, on all of these metrics. So just in terms of top line, we're already demonstrating, like, to the companies that we're working with, that we're increasing top line significantly as a result of our product. Right? So it's paying for itself and and much more, being able to take on additional demand, demand that's coming in that was never taken up before. Every company we speak to has, like, a a graveyard of leads that never got touched because they had no capacity to call. There's, like, a gold mine just sitting right there. So many of our first engagements is just unlock value from there. Right? And you just call, get the value from there. So I mean, there's

Utsav Somani: enrichment as well, like, on retargeting?

Raoul Nanavati (Navana.ai): Yeah. Yeah. Absolutely. So the like, every use case that we go into so let me talk a little bit more about that. Right? It it is not a product that we sell and say, hey. Here's the agent. Go ahead. Run it, and and, you know, you go forward. We have forward-deployed teams that are assigned to this customer for these business outcomes. They're tracked on a monthly basis. Context on a per customer level is tracked. So if I've called you last month, or, sort of, like, three times, and you missed your, and you didn't pick up and you missed your EMI. Right? It'll call back and say, hey. We tried you three times last month. You missed your EMI. Here's what happened to your credit score. Please make your EMI payment this month. Right? And, you know, we're happy to call you back. Do you need a reminder? So on and so forth. Right? And, you know, then we we can track that behavior across the rest of the organization as well. If there's cross-selling that needs to happen, that happens as well. I mean, they're the and we're just scratching the surface today. Right?

Utsav Somani: As a final closing one, I designed decided to DM you when I saw this, quote tweet that, Ronnie had done when I think Rajan Anandan was posting about Sarvam and the price per minute for a voice model. Yeah. Yeah. And there was some two sides of the argument. Can you summarize that for us in 20–30 seconds? What were the arguments being made on both sides?

Raoul Nanavati (Navana.ai): I think, Rajan was saying that if you're paying more than ₹2 a minute, that it is, you're overpaying. I think Ronnie was pointing out that look like, you know, whether it's 2 rupees or ₹3 or ₹4, is not as relevant as, building this out as reliable infrastructure. Like, being able to do that in a way that matters, like, on-prem that is secure, where data is not moving out, it is being done in a manner that is useful and not kind of, you know, spammy for lack of a better word. I think that was what Ronnie was pointing out that, like, ₹2 is not is not the goal here. We've already crossed the unit economic kind of, you know, argument. Like, we're already there. So pushing it down even further is not not necessarily the answer. Doing this in a way that is reliable, sustainable, private, secure, sovereign, this is what he was talking about.

Utsav Somani: Awesome. Thank you so much for coming on our show. Go, Navana, baby.

Raoul Nanavati (Navana.ai): Yeah. Thanks a lot, Utsav. Thanks, Dhruv, for having me.

Utsav Somani: Cheers. Got the CTO of super.money next. So my first question to him, Kaushik, it's gonna be which voice AI models are you using? There there are

Kaushik Mukherjee (super.money): a few that we continue to, work with. We'd love to, try Navana as well at some point. Why not?

Utsav Somani: Boom. We're making connections on Deepgram.

Raoul Nanavati (Navana.ai): I'll be in touch.

Dhruv Sharma: I was actually gonna say, like, banks went from branches to net banking to mobile apps to now voice agents. Who knows? Maybe voice agents is how market share is gonna be won and lost and, hence, you know, all of the love that companies like Navana are getting from banks.

Utsav Somani: Amazing. Yeah. I mean, that's actually true. So let's let's ask Kaushik that first question. I think everybody already knows super.money is a part of the fintech arm of Flipkart. After you guys spun out PhonePe, you acquired a bunch of companies as well, and I think you're the fifth-largest UPI app now in the country.

Kaushik Mukherjee (super.money): That's correct. We are the fifth largest UPI app.

Utsav Somani: So talk to us about the NPCI's agentic rails, which now they've introduced as well, for digital gold buying, I believe.

Kaushik Mukherjee (super.money): There are various use cases. I think, there are certain rails that have been introduced right now, but it's still, going through, some levels of experiments before we have enough confidence to, you know, roll it out in a larger scale. But, yeah, voice is, one of them, where we are sort of, having some guardrails with which we're sort of rolling it out currently itself.

Dhruv Sharma: Gotcha. Give us a sense of how far embedded finance has come from, you know, in early experiments to today's deployments.

Kaushik Mukherjee (super.money): I think, it's been quite a journey, I would say, in in terms of, the, the kind of progress that's happened, when it comes to finance. I think, not only from a from a technology standpoint, but also adoption. Right? I think, it's become a lot more, how should I put it? I think, it's it's it's a lot more customer centric now. It's a lot more intuitive, I would say, and a lot more commonplace.

Utsav Somani: And as your role, I mean, your role CTO, right, I mean, of super.money, you're processing over 200 million transactions a month, I believe. And, I mean, the volume is tremendous. The rails are very common. So what do you build on top to increase reliability and selling credit products? Like, what does your role typically involve on a rail which is publicly available to everyone at the same time? Yeah.

Kaushik Mukherjee (super.money): So yeah. So it's close to 450 million a month, by the way. I'm just letting you know.

Utsav Somani: It was offered to X.

Kaushik Mukherjee (super.money): Yeah. And and, it sort of we we we continue to grow. I think the first thing and and and and the most important aspect, of service to ensure that, you know, there is consistency and trust, as far as the platform is concerned. That's UPI is the very base offering on top of which you then end up building various kinds of cross-sell products. But your base offering needs to be, you know, absolutely kick ass. Right? What that means is you need to be available. You need to be consistent in terms of the experience. You need to build trust. I think those are some of the, pillars on which, you know, we look to then figure out how to build verticals, if you may, that are intuitive, as far as the end customer is concerned. Hence, very deep hyper-personalization across these verticals. And how do you achieve that at that kind of a scale, would be, you know, the problem statement, I would say. Yeah.

Dhruv Sharma: So big billion days, less than three weeks, or roundabout that far, Kaushik. Is that, like, is that a very busy time of

Kaushik Mukherjee (super.money): the year for you guys? It's always busy. It's always busy. And, you know, the, the food ordering is up at night. There's, a lot more Swiggy and Zomato orders, going on and, a lot more than usual. Yeah. It's it's that time of the year.

Utsav Somani: You're, I mean, CEO of the fifth-largest UPI app. I think we have to ask you the UPI MDR question. Were you guys celebrating when the news came out? What was the first reaction of the team? Like, I mean, was it a sigh of relief? I mean, what was the feeling like?

Kaushik Mukherjee (super.money): I mean, it's good news, in terms of the fact that, there is obviously, you know, that there is going to be a certain amount of revenue that's getting generated. But even otherwise, I think that, from our end customer standpoint, we don't necessarily see this creating, you know, any, other impact. So it's a win-win, for all is what I would say. Yeah.

Dhruv Sharma: And there's been all of this talk around agentic commerce, Kaushik. What's it looking like in production and, you know, what are the sort of backroom conversations going on right now? Yes. A lot

Kaushik Mukherjee (super.money): of experiments. We we recently launched split store. It's essentially, you know, you know, buy now, pay later kind of an offering. And it's taken off, right from the time we launched it. Huge amount of adoption. I think, from from an agentic standpoint, intent discovery and, and and sort of figuring out, what, you know, would make sense for this particular customer. Can an agent have, and some of these are still in the experimentation phase. Can an agent do the purchase for you once it meets certain kinds of price points, are, you know, things that we sort of, deliberating on. So yeah.

Utsav Somani: 400 million is a massive transaction volume. Like, I mean, just in terms of sheer amount of transactions going through, your team has described Sentinel, which is an internal anomaly detection system. So talk to us about how you handle fraud at scale.

Kaushik Mukherjee (super.money): Yeah. You know, I'm I'm I'm going to reveal as much as I can because it's fraud at the end of the day.

Utsav Somani: Yeah. You don't want, like, the people on the other side to become smarter by realizing how you're handling it. So, yeah, I mean, feel free to, of course, exercise it.

Kaushik Mukherjee (super.money): Trust me. They're trust me. They're very, very smart. Okay? I mean, this is one thing that I'll sort of give to them. But, I think an anomaly detection essentially is, this tenet of, any deviation from the norm. Right? And how do you, figure that out at, our scale? We're talking about close to I mean, we were at a billion events maybe last month, and that's grown even further. How do you sort of, anticipate? How do you classify, certain kinds of anomalous behaviors in, different kinds of flows? How do you figure out if something is indeed anomalous? Because you don't want to penalize a good, you know, user, at the end of the day. So, it's a fine-tuning, problem, if you may.

Dhruv Sharma: Can you also talk to us a little bit, Kaushik, about, like, what it run what it takes to run, you know, a hyperscale payment system? Like, all of this money is moving. The UPI architecture's got several parties. You know, some banks are more prone to outages than others. You know, how many requests you receive, how do you route them,

Kaushik Mukherjee (super.money): all of that. Yeah. Yeah. So, at at the heart of it is resiliency of as far as your platform is concerned because, you know, the call graph is quite extensive from the time a request is generated till the time it gets fulfilled. You have a lot of parties through which that packet sort of traverses, if you may. So, you as a platform need to be resilient. You need to sort of have some early warning signals in case one particular partner is going through, you know, some kind of an availability concern. And you need to be, intuitive in terms of your communication in the event that a particular partner is going through, let's say, their own challenges. Right? So that, from a you you you essentially are trying to obfuscate the complexity, and you're trying to provide information, which is a lot more commonplace, a lot more intuitive as far as far as the end customer is concerned because they don't care whether it's, you know, NPCI or super.money or some particular bank or what the five x six was. They really want to know what happened to their money. Right? And you have to be, as simple about it as possible without being wrong, think. That's important.

Utsav Somani: Is that why you diversify your banking relationships as well? Because I read that, you have bank, I mean, sorry, FD-backed credit cards as well as prepaid instruments. So do you I mean and partnerships with three banks, I think, for serving that product. So is that why you differentiate or diversify your relationships with banks as well?

Kaushik Mukherjee (super.money): Not really, but but that's a use case I'm happy to talk about. You know, one of the, one of the, areas where we believe, as an organization that we, can make a difference on the kind of, risk data that we, that we capture, the kind of risk analysis that we do. And we believe that we would be able to provide financial instruments to anybody who's aspiring to have a financial instrument. So in that context, if you were to look at, let's say, the FD-backed card, that's a card which is given to folks who may or may not have a CIBIL or an Experian score. Yeah. Card at a particular score, you end up being eligible for a card. Right? But then that doesn't necessarily mean that the person who does not have the score, is does not have the right to have a card. Right? So so so so in that context, what we did is we ended up building this FD-backed secured credit card where you put in some money that goes into FD, and we, in in lieu of that, give you a card. Over time, as you end up using the card and you start showing the right kind of behaviors, your Experian score actually gets better. Right? Now, and and what that does is then it makes you a lot more eligible and, towards unsecured cards, like a like, maybe a Axis Bank card that you might have seen on our platform as well. Right? So, essentially, what we are doing is we're trying to democratize some of these financial instruments for all and sundry, and going along with the journey of people who are aspirational and are showing the right behaviors. Right? That's

Dhruv Sharma: Are you already starting to see signs of a future where people don't even come to the primary product surface to to shop, to browse, to make a transaction? They'll just use their personal AI agents. I think, like, so much No. I instinctive about all of that.

Kaushik Mukherjee (super.money): I have a point of view on that. I think, see, there are certain kinds of use cases where, for example, we were talking about voice a while back. But, you know, voice also has and it's of it, not necessarily having the right kind of privacy. Like, for example, when you think about tapping and buying certain items, you it's, you know, the the the the kind of privacy is encompassed within how you are consuming that content and what you want to buy versus, let's say, unless you're in a confined environment. Now now in that context, there will be certain kinds of use cases to your point, Dhruv, where, I don't necessarily see that people would always want to get into this particular destination to actually achieve what their means are. But then they'll all but it's compartmentalized. Meaning, there'll still be use cases where where you would want to go to that destination, be it research, be it privacy or whatever. Right? Towards, yeah, so, not necessarily a zero-one answer in that sense.

Dhruv Sharma: Yeah. The Silicon Valley echo chamber is talking about the death of the UI and the in Korea, they have those dopamine apps where you can keep browsing and not make a single purchase. So Yep.

Utsav Somani: I feel that. I think Dhruv, you must be referring to that single tweet. Right? On that x account where he said that now he doesn't log in to his x, Gmail, Google Sheets and stuff. I think the chatbot in itself has become the interface for them. But the counter side that I I read from one of the comments is that for some people, like, browsing the Internet, which is, like, I mean, window shopping of an ecommerce website or a travel portal and that they're dreaming that they get to do while doing that or making the purchase. Individual purchase, of course, not an enterprise purchase. I think that's the joy, or more than the purchase.

Kaushik Mukherjee (super.money): 100%. I think I think and and and that's an interesting point because, you know, the way my daughter by the way, she discovers some of these lifestyle items is when she'll go to Instagram and circle certain items, and then maybe the that particular product is available on Myntra. Right? And that's how she perhaps discovers Myntra. But that use case still continues to be there where you there are going to be certain kinds of use cases where you still want to get to that destination site.

Utsav Somani: But have you used Muse and Instinct? Apparently, the invites are now floating around in India as well?

Kaushik Mukherjee (super.money): Yeah. I did. We did use so one of my, one of, one of the folks in office pinged me saying the Instinct, computer running really, at capacity, and they're adding things manually. So not much agentic there, I guess. But, yeah, used Instinct recently. Pretty cool. Yeah. Good stuff.

Utsav Somani: Nice.

Dhruv Sharma: How how has the team changed from the inside ever since the AI has gotten better and better and better, Kaushik? How are you guys interviewing? Are you hiring a lot? What's happening?

Kaushik Mukherjee (super.money): Yeah. I mean, we, so we we had the advantage. I mean, at least, super.money had the advantage of not having to deal with too much legacy systems when we're building because we built it in in the AI age, so to speak. Right? So it's not as if we ever overhired. We are pretty nimble in terms of how we get things done with the team size that we have. But our interviews, when we when we want to get somebody on board, I think it all it also sort of hinges on to what extent this person really understands the utility of AI. Our SDLC is significantly, I I I think most of our SDLCs today, driven by AI, not just code generation. I'm saying, right from the time that you we write code to deployment, CI/CD, auto-rollbacks, etcetera. Incident management, a lot of that is via AI. Provisioning is via AI internally speaking because you asked that general question.

Dhruv Sharma: Yeah.

Kaushik Mukherjee (super.money): So a lot a lot of, areas where, you know, you have repetitive constructs or for that matter, pattern matching. A lot of that is AI. But, architectural conversations, design what what is a good design, is still a lot more human, from a software standpoint at least. Yeah.

Utsav Somani: Kaushik, because a final one, for all the aspiring CTOs that are listening to this, they wake up like you, suppose, within a Flipkart group company as a group CTO of one of the largest payment apps in the country. You're, of course, worried about reliability. You're worried about fraud. What are the things that get you excited?

Kaushik Mukherjee (super.money): I think the I there is a you know, that's a great question. One of one of the things that we are, really, banking on is, some of these lines are getting blurred. Meaning, we just launched split store. Now how does this construct of UPI the strength of UPI is, you know, the high amount of engagement that UPI has. Right? How does this construct of UPI lead to higher amounts of cross sell, on an on a commerce entity. Right? Then we have then we perhaps have an answer, for, you know, engagement that goes outside of, let's say, an Instagram or a Facebook. Right? When we're when we're thinking about, doing purchases. Right? Or for that matter, to point, some of the other modes through which discovery happens, as far as some of these product listings are concerned. And, you know, the opportunity to actually experiment on some of these things with the front row seat is what excites me endlessly, I would say.

Utsav Somani: Amazing. Wishing you the best, and keep creating impact. Thank you so much for coming on our show.

Kaushik Mukherjee (super.money): Awesome. Thanks, Thanks,

Utsav Somani: Alright, listeners. Moving on to our third and final guest today, Priyadarshi of CureBay. Priyadarshi, welcome to the show.

Priyadarshi Mohapatra (CureBay): Hi. Thank you so much for having me on the show.

Utsav Somani: Where are you dialing in from?

Priyadarshi Mohapatra (CureBay): Bhubaneswar. Sure.

Utsav Somani: Oh, nice. So let's start with an introduction to CureBay. What do you do at CureBay?

Priyadarshi Mohapatra (CureBay): We try to solve for last-mile trust in rural healthcare.

Utsav Somani: Expand that a little bit.

Priyadarshi Mohapatra (CureBay): Alright. Take two weeks. Alright. So we realized that all the healthtech that has been built in this country is very urban-centric, is very top-eight-city, capital-city focused. That caters to 300–350 million people. About close to a billion people living in what we call as rural India, Bharat, and many other names. And the only infrastructure for health care that's available to them is what's built by the government. The PHCs, the CHCs, the district headquarters hospitals. And government has done an amazing job in building an infrastructure, but as a country, we're challenged to have providers there. So access to health care has been a challenge, and that's what we're trying to solve.

Dhruv Sharma: Is there a human element to deliver delivering health care, in rural India, Priyadarshi?

Priyadarshi Mohapatra (CureBay): Oh, absolutely. You know, this is not something you can solve by building an app or a tech platform. Unless that rural patient has experienced somebody putting a stethoscope doing auscultation or measuring his blood pressure, this happening. Mhmm. K. So the entire approach is a very hybrid approach, because eventually, it's it's not the, clinical part that's winning the game. It's really the trust part, the empathy part that's winning the game. So you gotta get the human in the loop there. You gotta get the right full-stack health care at the doorstep, and that's how you sort of change behavior.

Utsav Somani: And tell us more about these circles. How do you get trained health care staff, medical staff to these places, these remote places? And what are eclinics?

Priyadarshi Mohapatra (CureBay): Yeah. So so that's interesting. That's the core of our model. So, you know, what we're really trying to build is what we call as probably the world's first rural health care operating system. And our operating system actually has multiple layers. So the first layer is our network of clinics. These are physical clinics, about 200, two fifty square feet smart clinics that are in the remote area. But each clinic has a human element, which Dhruv was referring to. Right? So we have a trained nurse and a trained pharmacist who are physically present in our clinic. They have access to a tech platform which we have built. The doctor, the hospital, the diagnostics all sit on our platform. So you have a human provider inside the clinic. They do your auscultation. They measure your blood pressure. They hold your hand, and that's when empathy translates into actually helping people heal. And then you pull in the doctor, you pull in the other services that is required, and hand hold them through this health check. So we have had challenge in trying to find the right kind of people, but we have now built that model on scale in a centralized model where we recruit people, train them, and try to place them closer to the place they belong to. And that's what has really worked for us.

Dhruv Sharma: And can you give us a sense of, when people walk into these clinics, what do they come looking for? Is it just consultations, or is it post-operative care? Give us a sense of all of that too.

Priyadarshi Mohapatra (CureBay): You know, Dhruv, I'll be absolutely honest. Initially, when we opened the clinic, we thought we are gonna have a queue outside our clinic because they don't have access to health care. And here, we bring in everything, and we were surprised to see people not coming in. And then when you go and speak to them, you realize that, the way rural India seeks health care is very, very different than how you and I see. And I keep saying we are privileged. Something happens to us, our family, our peer group, everybody will tell us, listen. Take a day off when meet the doctor. For them, it's a constant debate because it's one day loss of income. So can they procrastinate it by a day? Right? They will try to try to handle it at home. And when they're coming to you, the condition is already acute or chronic, and they really need health care. So we encounter a lot of people walking into a clinic who are in dire need of immediate health care. So we get them to speak to a doctor. We get them to speak to a specialist. Our people get the tests that are required to be done there. And if required, if there's an intervention requirement is there, we send them to a partner hospital. But, very, very interestingly, we have seen also people just walk into the clinic saying because it's an air conditioned place. In the heat of the place that we are operating in, which is the eastern part of the country. Right? They sometimes come just sit in the clinic, have a glass of water, and, we believe that that's also health care because you're healing somebody. They're feeling good about it, and they're going out. So so we've had very different kinds of people walking into our clinic.

Utsav Somani: And trust I mean, so brand building, of course. I mean, in our major tier-one cities, we think about Instagram ads and all of these things. But in trust and brand or the trust towards a brand in rural areas must be very, very different, especially in in terms of health care. So how do you I mean, the small, small things that you do while you set up these eclinics, what are the things that help build that trust and enhance it?

Priyadarshi Mohapatra (CureBay): Notes are bang on. I mean, you don't solve for rural health care through tech or building clinics. You solve by building a trust infrastructure, and for us, it's in two layers. One, you gotta connect with the community. So we actually take people from the community whom we train as healthcare champions, empower them with an app which sits on their phone where all the services of the clinic can now get extended to the doorstep of people and the community they belong to. And these are people who have the tribal knowledge. Right? They know exactly who's expecting a baby, who slipped and broke their leg, who's been down with fever. So they go and tell them, listen. I will get the doctor to speak to you on my phone. I will get somebody to come and take your blood samples. I'll get the medicines delivered here, and that's the first level of trust you build. The second is the fact that we have the clinics, and it's permanent in rural India has actually become a little bit sick and tired of urban experiments. They actually tell you, you know, we have these lot of people coming and doing camps, but they come into camps at their convenience, not when I need health care. So when you have a clinic, they know that when they need health care, they can walk into your clinic. That's the second level at which you build trust. And I keep saying I said the way we're trying to solve this is by building that trust infrastructure. Everything else follows.

Dhruv Sharma: Do you see at least a subset of, patients, maybe the older generations, who view private health care with a healthy dose of skepticism?

Priyadarshi Mohapatra (CureBay): Well, actually, you'll be surprised to see that, they're very, very open to it. And, you know, it's bit of a myth that they don't come. The only reason I would say the holdback is because of that debate in loss of income. Because for them, it's the cost of health care is really not the cost of that consultation with the doctor. It's the cost of missing one day's income. It's that bus ride cost that they will take. It's the cost that they will spend because they have to spend on all day by going to the closest place where health care is available. That's really the cost which compounds on them, and that's where for the whole bag. Otherwise, they know. They're very aware today. Because there'll be some youngster in the family who's told them that, listen. This informal quack system doesn't work, and it it's not the best for your health. You need to go and see a doctor.

Utsav Somani: And you acquired a pharmacy distribution business, which is far away from the clinics that you run. Why that, decision?

Priyadarshi Mohapatra (CureBay): So we were very clear. If you have to meaningfully solve for health care, you will solve for full-stack. So telemedicine on its own really doesn't work in rural India. Right? Because, it solves for the video call. What happens after that? A doctor generates a prescription, and you need to close the loop. All the

Dhruv Sharma: call to action in the

Priyadarshi Mohapatra (CureBay): prescription, whether it is prescription, whether it is medicine that is required, whether it is diagnostic test that is required, needs to be solved. And while we were trying to solve that, we realized that the distribution of pharmacy in rural India itself is extremely challenged, because most of the distribution are very, very, you know, urban-centric. So to ensure that even the ecosystem has the right medicine so people can go and get that, we had to build it from scratch. So we acquired a business which was a tech-oriented business. So it sat very well with our own platform, and we could really facilitate the right kind of medicine reaching to the remotest parts of the country.

Dhruv Sharma: Can you talk to us about how you're thinking about, scaling investment, profitability?

Priyadarshi Mohapatra (CureBay): So let let me take a step back and tell you, Dhruv, you know, when I quit Google to start this, I was very clear that, you know, this is not something build, which is gonna be based on CSR grants because that's neither sustainable nor scalable. If you have to meaningfully solve for this, you gotta create a model, which is commercially attractive because it's a non-trivial problem. And to solve a non-trivial problem, you need non trivial capital to follow. So you need the right commercial capital to back you up. So we were very clear that the unit economics has to make sense. It has to be a viable model. So there is no free health care that we offer. So we make it extremely affordable. Like, you can walk into a clinic. You can do a consultation first consultation for ₹100. Then you get the medicines test, everything which is of the highest quality but below the market price. So affordability is what we stress on, but it's a service that you pay for. And only then, you also value the service. And that's how we've sort of grown. We look at a approach called as a circle where we take about 50-odd clinics because you need to build a rider, network because there's nothing really existing there to help you sort of facilitate the movement. So as you build this, you need a minimum number of clinics to make it commercially viable, and we today operate in Odisha and Chhattisgarh and we had about 240 clinics and growing strongly.

Utsav Somani: Priyadarshi has a final closing one. You mentioned your ambitions to go full-stack in health care. What is something that you will not do? So the complete opposite.

Priyadarshi Mohapatra (CureBay): Probably, we will not get into opening hospitals for now. Right? Because there is an ecosystem of hospital partners who are there, and we really want the ecosystem to come and participate in this opportunity. You know, again, going looking at my background, I think when I looked at Microsoft and Google, I realized that platform wins. So you need to create the right kind of platform where the ecosystem can come and participate to it. So the parts of the ecosystem that doesn't exist we are building and the parts of the ecosystem like hospitals and all, which exist at a certain place, we are connecting them to reach the last mile.

Utsav Somani: Awesome. Wishing your best on this mission. Thank you so much for coming on our show.

Priyadarshi Mohapatra (CureBay): Thank you so much, Utsav and Dhruv. It was a pleasure talking to both of you.

Utsav Somani: Thank you. Alright, listeners. That's it from us. Have a safe and a fun weekend. We'll see you on Monday for stream one four one. Thank you. Good night.