K. Vignesh, HyperVerge |“ A 500 line code is small for me, but means a lot for another person"| Ep.5

Speaker 0:00 Hi, my name is Amrit. We've heard that IIT Madras is the best place to build. [upbeat music] So we've come down to the Sudha and Shankar Innovation Hub. We want to meet some people. These are builders. We want to talk to them about their work, and also ask them, "What makes IIT Madras the best place to build?" [upbeat music]

Vignesh Krishnakumar 0:28 But just the thought that, you know, something like this could eventually solve this problem created so much joy for him. So I tell people that if anyone else had been in my situation, they would have wanted to start a company, too. Post 2019, if you have taken a SIM card, chances are that you would [chuckles] have gone through HyperVerge KYC systems.

Speaker 0:49 [upbeat music] Hi, welcome to The Best Place To Build Podcast. Today, we are sitting with K. Vignesh, or KV, as he's known. Uh, KV is the CTO of HyperVerge, one of India's largest AI companies. You may not have heard about them because their product goes into other products. So let's start with welcoming Vignesh. Hi, KV.

Vignesh Krishnakumar 1:18 Hi, Amrit.

Speaker 1:19 Uh-

Vignesh Krishnakumar 1:19 Thanks for having me on this podcast, and always happy to be here. Um, the Centre for Innovation and IIT Madras is home to us. This is where everything began for us.

Speaker 1:30 Right. You have a deep connection with CFI. Uh, we'll get into that, but let's do one thing.

Speaker 1:35 Let's start with what is HyperVerge? What does it do?

Vignesh Krishnakumar 1:37 HyperVerge is an AI platform for banks, financial institutions, and other large enterprises to verify and onboard their customers in a remote and automated way.

Speaker 1:48 Right.

Vignesh Krishnakumar 1:49 In many parts of the world, this is referred to as a KYC process. So what is KYC? Uh, in simple terms, are you actually who you say you are, right? And are the credentials that you're providing verified and actually belong to you? For example, your mobile number, your bank account, your address, are these details actually authentic, and do they belong to you?

Vignesh Krishnakumar 2:14 Why is KYC, uh, important, right? Uh, now, the first factor is, uh, trust in any day-to-day interaction comes from verifying the person who is performing the task. Imagine, uh, if I take a loan in your name-

Speaker 2:31 Sure

Vignesh Krishnakumar 2:31 ... and I take that money and abscond. Someone comes knocking on your door to recover the money, uh, but the money would've been with someone else, right? So verifying in a really reliable manner that you are the person who is applying for the loan ensures that the money is safe and can be repaid appropriately, right? Similarly, if you take opening a bank account, verifying the bank account holder ensures that the account is safe, the account is not used for tax evasion or money laundering or funding terrorist activities or other antisocial activities, right? Similarly, if you take someone getting a SIM card, only if you can verify who is the individual who is taking the SIM card can you ensure that the SIM is not used for unlawful purposes. So this is where KYC comes in. It ensures that trust in day-to-day interactions is something that is taken care of. The second reason why, uh, automated KYC is helpful, if we look at, uh, the, the way it was done before, uh, say, if we take, uh, an affluent person in the city, uh, for example, uh, and you want to open a bank account or you want to apply for a loan or invest in a product, something like that, uh, you can walk into a branch or a relationship manager or a branch correspondent will come to your place. Uh, they will bring in the forms, you can fill it out. They will take, uh-

Speaker 4:02 Right

Vignesh Krishnakumar 4:02 ... copies of your, uh, ID cards and other documentation.

Speaker 4:06 I mean, I'm not an affluent person, but that's happened to me also, where the bank guy has come to my house.

Vignesh Krishnakumar 4:10 Has it? You might consider yourself not affluent, but-

Speaker 4:13 Somebody else might. [chuckles]

Vignesh Krishnakumar 4:14 Yeah, exactly, right?

Speaker 4:15 Yeah.

Vignesh Krishnakumar 4:15 Uh, so they will submit all of these details. Someone, uh, verifies the details, uh, does data entry, ensures that the person whom- who was meant and the person whose details are submitted are the same.

Speaker 4:27 Same person.

Vignesh Krishnakumar 4:27 The documents are verified, all of these things, right? Now, when this happens in this manner, the cost of processing this application is significantly high. The cost of-

Speaker 4:37 The bank is willing to take that cost from me.

Vignesh Krishnakumar 4:38 Willing to take the cost from you. But if we go, say, 300 kilometers outside the city, the need for essential financial products remains the same, right? The need for savings, the need for credit, the need for-

Speaker 4:53 Insurance

Vignesh Krishnakumar 4:53 ... investments-

Speaker 4:54 Yeah.

Vignesh Krishnakumar 4:54 - insurance.

Speaker 4:55 Yeah.

Vignesh Krishnakumar 4:55 These are essentially the same for that individual as well, right? But the ticket size or the amount for which they might be taking that financial product might be very different.

Speaker 5:05 Sure.

Vignesh Krishnakumar 5:05 And the cost of processing that application might be significantly higher and might not be viable at that ticket size, right? Imagine you take a home loan or you take a personal loan for lacks of rupees. Someone else in a village might be taking a loan for a 3,000 rupee or a 5,000 rupee, and, and so on.

Speaker 5:22 So the margin is too small for me to send someone to verify-

Vignesh Krishnakumar 5:25 Exactly

Speaker 5:26 ... he stays where he's saying, if it's the same person.

Vignesh Krishnakumar 5:28 Correct. So when these checks are done by AI in a fully automated manner, where people sitting anywhere can just click a selfie of themselves, can click photos of their ID cards and other documents, submit it, and this whole process is verified using AI, then the cost of processing this application goes down drastically-

Speaker 5:49 Right

Vignesh Krishnakumar 5:49 ... that the financial product becomes more affordable to everyone in the country, right? Second, it becomes accessible anywhere, any time. If I'm in the middle of the night, I have a, you know, urgent requirement for some financial product, say, a small ticket loan that I need to apply for-... I can just download an app from a play store and do the process, be done with it, and have money even disbursed into my account in less than five minutes. The accessibility of the financial product also increases when this KYC verification is done in a remote and automated manner using AI.

Speaker 6:24 I remember speaking to you, uh, some time back, and you were telling me of a project where, uh, HyperVerge is also helping, uh, check for somebody's liveness.

Vignesh Krishnakumar 6:33 Yeah.

Speaker 6:33 Uh, is that also- so, like, I- correct me if I'm wrong, the use case was for people to prove that they are still alive so that they can collect their pension. Is that right?

Vignesh Krishnakumar 6:43 Yeah, that is also one of the use case. So, uh, identity verification is applicable in a lot of, uh, processes, right? One such example is disbursement of pension. Uh, again, here, previously, if, if you had elders in the family who collect their pension, they will have to go to a branch in person to prove that they are alive, and this is the person who is actually the, uh, pension account holder, right? And post this verification, something known as a, a liveness certificate is issued, and basis that, their pension is disbursed. And this is something that they have to do every year or at some, uh, particular frequency, right? Now, with AI, you can be sitting at your home, you can capture a selfie or record a video, and from that, we can verify that there is a live person, and this matches with the details of the account holder, and the live certificate is issued in an automated way, and the pension is disbursed, as well, automatic.

Speaker 7:43 So what are we- so what you're saying is that the financial institution or, or the o- or a telecom institution needs to know who you are?

Vignesh Krishnakumar 7:50 Yeah.

Speaker 7:51 Needs to be able to trust your, uh, identity.

Vignesh Krishnakumar 7:54 Correct.

Speaker 7:54 And there, if it's a remote operation, and I ha- and I can't afford to send someone to your place, then a remote KYC-

Vignesh Krishnakumar 8:02 Yeah

Speaker 8:03 ... uh, sort of products like HyperVerge comes into play. So I wanna ask you a question. So today it feels very obvious, because, like, my phone knows who I am because I can scan my face and it opens up, and it does so, like, you know, in a fraction of a second. I just open it, and it's opens, right? So you started way long back, right? Like, I think, uh, iPhone introduced face unlock in 2017.

Vignesh Krishnakumar 8:25 Yeah.

Speaker 8:26 So y- you are predating that.

Vignesh Krishnakumar 8:27 Yeah.

Speaker 8:28 So it's interesting, and also,

Speaker 8:30 mm, I saw on your website, I've heard you say- talk about it, that in India, almost 300, 400 million pro- people... Uh, correct me if the number is wrong.

Vignesh Krishnakumar 8:39 Yeah, more than that right now. [chuckles]

Speaker 8:40 More than that have used HyperVerge, uh, face match-

Vignesh Krishnakumar 8:45 Match

Speaker 8:45 ... liveness and, uh, fraud detection, right?

Vignesh Krishnakumar 8:47 Yeah.

Speaker 8:47 So if I'm a customer, and I'm listening to this show, uh, huh, because your product goes into something else, so how will I know if I've used HyperVerge?

Vignesh Krishnakumar 8:55 Yeah. By now, more than a billion people have gone through HyperVerge identity verification, uh, systems in one form or another. The problem that we spoke about is universal. Uh, it's not just in India. Anywhere else in the world, if you have to open a bank account or buy an insurance or invest in a product or get a loan or, uh, purchase a SIM card, um, the, the process is the same, right? And there will be local variations, local regulations that are different, but largely the need to identify who is the individual who is, uh, receiving this service is the same. So more than a billion people across the world have gone through this process.

Speaker 9:39 That is amazing.

Vignesh Krishnakumar 9:40 Uh, talking about examples in India, two of the three telecom companies in India use HyperVerge for identity verification for their SIM card disbursement process. So post-2019, if you have taken a SIM card, chances are that you would have [chuckles] gone through HyperVerge KYC systems.

Speaker 9:59 And what changed in 2019? Why did you say post-2019?

Vignesh Krishnakumar 10:02 Around the end of 2018, uh, Supreme Court had come up with this, you know, uh, guideline that, uh, private, uh, institutions, um, should not use the other biometric system, fingerprint-based biometric system, for disbursement of, uh, their products for, for KYC. And at that time, telecom companies were amongst the largest consumers of this biometric eKYC, uh, system. Jio was growing, uh, rapidly. Um, and, uh, at that time, Jio was onboarding around a crore customers, uh, each month. One crore SIM cards were being disbursed each month, and this regulation had hit them very hard. Their onboarding rate had drastically dropped from, you know, crore customers to significantly lesser, and they had resorted to a manual process, verification process, where, uh, people had to, you know, walk in with their ID card, uh, the customer photograph was taken, photograph of these documents were taken, and someone was verifying a lot of these details in the back end.

Speaker 11:08 So earlier they were using Aadhaar. Supreme Court said, "Don't use Aadhaar, because you're a private institution. You don't have the... I don't know what you're going to do with Aadhaar data, so don't collect Aadhaar data." And then, uh, they said, "Oh, now we can't use Aadhaar. We have to do everything manually."

Vignesh Krishnakumar 11:20 Yeah.

Speaker 11:20 And India, of course, is a really large country, which would mean, like-

Vignesh Krishnakumar 11:24 Okay

Speaker 11:24 ... a million people-

Vignesh Krishnakumar 11:26 [chuckles]

Speaker 11:26 ... taking photos at-

Vignesh Krishnakumar 11:28 Yeah.

Speaker 11:29 So then they moved to, uh, digital onboarding?

Vignesh Krishnakumar 11:32 Yeah, Hyper... So there are multiple challenges existed at that point of time. One, your turnaround time for processing the application increases significantly, right? So, uh, the way SIM card onboarding happens is you walk into a store, you submit these details, and you walk out with a physical SIM card.

Speaker 11:49 Yeah.

Vignesh Krishnakumar 11:50 And post the verification of your identity, the SIM card gets activated, right? Which means that if the agent who's present with you has failed to capture details properly, um, or your, uh, document-

Speaker 12:04 Just typed my name wrong.

Vignesh Krishnakumar 12:05 Or, uh, has typed the name wrong, or your selfie is not clear, or-... uh, there are multiple people standing in the background, and you don't know, like, whose photo this is. Anything that is wrong in the application.

Speaker 12:17 Yeah, sure. Three people have come, I have taken your photo and taken his ID card-

Vignesh Krishnakumar 12:20 Exactly

Speaker 12:20 ... what if it's, like, by mistake?

Vignesh Krishnakumar 12:22 Right. What ends up happening is that the, the, the user walks out with the physical SIM card, but the application post-review gets rejected later, right? And a physical SIM card is lost. It can't be activated, uh-

Speaker 12:36 And it's a very bad customer experience.

Vignesh Krishnakumar 12:37 It's a bad customer experience, right? Second, the turnaround time of this application increases drastically from, uh, them being able to, uh, activate the SIM card in less than five minutes to, on an average, it increasing, uh, beyond 60 minutes, and in some cases we've seen it go till 12 hours or so also, right? Which means you have unhappy customers who are calling their customer care and service centers and asking, "What is the status of my SIM card activation?" Right? "Is it done? When will it get activated?" And because of that uncertainty, uh, they keep getting, uh, constant calls. And, uh, when people are unhappy, it is very easy for them to switch over to some other provider who's able to activate faster, right? So this was a problem that, uh, they were facing. So we created an automated KYC process leveraging AI, where, uh, every check that needs to be done on the customer's photograph, on the documents that are submitted, are done real time, instantaneously, and the feedback is given to the user, right? And if everything passes, your SIM card activation also happens before you walk out of the store itself.

Speaker 13:47 Right. So in 2019, the telecom companies at the time moved to HyperVerge?

Vignesh Krishnakumar 13:52 Moved to HyperVerge for their onboarding process. Other examples-

Speaker 13:57 So what you're saying is, if I have bought a SIM card after 2019, [chuckles] then I have used HyperVerge.

Vignesh Krishnakumar 14:01 Very likely.

Speaker 14:02 Very likely.

Vignesh Krishnakumar 14:02 Very like. Uh, if you've opened a bank account with SBI, uh, digitally or taken a credit card from SBI, uh, very likely that you would've gone through HyperVerge verification process. Pension account holders on SBI, uh, could have leveraged the online, uh, verification process for pension disbursement. Delivery partner on Swiggy capturing a selfie to authenticate, uh, themselves, right? Uh, to know that the person who was originally verified for a person with decent background checks and police verification, et cetera-

Speaker 14:38 Sure

Vignesh Krishnakumar 14:39 ... is that the same person who is right now delivering things to you, right? Uh, someone is knocking on your door and entering and delivering a parcel, or as part of Swiggy Genie, they might be collecting your item, and they might be delivering it somewhere. How do you know, you know, you are safe or the item is safe? Uh, Swiggy uses HyperVerge for-

Speaker 15:00 Oh, that's interesting

Vignesh Krishnakumar 15:01 ... verification of the delivery partner at the beginning and authentication of them later, um, as they do their, uh, regular deliveries, right? Uh, that is another place where, uh, you might have seen HyperVerge systems.

Speaker 15:15 So what you're saying is that if I've used SBI, if I've used, um, uh, if I've opened a, uh, if I've taken a SIM card, if I have bought a financial products in the last four, five years, very likely that I've used-

Vignesh Krishnakumar 15:26 Yeah

Speaker 15:26 ... HyperVerge.

Vignesh Krishnakumar 15:27 Yeah.

Speaker 15:27 That-

Vignesh Krishnakumar 15:27 Investment is another example. Um, seven out of the top 10 investment apps in the country use HyperVerge for their KYC and verification process.

Speaker 15:37 I also understand that since your technology goes into some other companies, there must be companies you can't talk about because your, your NDA will not allow it, so let's just leave it to people's imagination. [chuckles]

Vignesh Krishnakumar 15:47 Yes.

Speaker 15:47 Um, but that's amazing. So a billion people across the world have, uh, used HyperVerge at some point. What I'm interested to know is, uh, you started before face match, so I think this is an interesting question to ask you. Uh, uh, sorry, but you started before face match became a commonplace thing. Now, it's obvious. There

Speaker 16:05 is this criticism of AI companies that most AI companies are wrappers, and that they use some API call, and they're just building a product around it. Is HyperVerge like that, or is it different?

Vignesh Krishnakumar 16:14 So HyperVerge builds, uh, all... I mean, almost all of our AI models are built in-house. Our primary focus is on problems that general purpose AI technology cannot solve really well, and I will explain this with examples on what we are doing. Um, so if you take face recognition as a general problem, right? Like you said, there are enough face recognition APIs out there. Every standard cloud provider, AWS, uh, Google, Azure, all of them have their own, uh, face recognition models, right? Uh, and there are lots of general purpose face recognition models out there. But if you take the specific process of KYC for a financial institution and the way that needs to work, and what are the consequences of that not working well, there are lots of nuances in that which needs to be solved, uh, much, much better than what a general purpose, uh, face recognition model can do, right? For example, if you take a match between a customer photograph and an ID card, right? So this ID card, uh, might have been taken at whatever age. Today, if you take an Aadhaar card, you might have taken it when you were 18 or before that, your driver's license, your voter ID.

Speaker 17:31 I mean, when I was 18, Aadhaar was not there, but-

Vignesh Krishnakumar 17:32 Aadhaar, yeah. So for, for a lot of people, uh, the, the ID-

Speaker 17:37 You've grown quite a bit.

Vignesh Krishnakumar 17:38 Yeah.

Speaker 17:38 Or your fa... or, uh, you know what happens to me a little bit sometimes, is that if I'm changing my specs, then it doesn't recognize.

Vignesh Krishnakumar 17:43 Exactly. So a lot of these changes happen, right? And in India, you have so many of these variations. How you looked earlier, how you look now might be different. There are so many ethnicities, so many hairstyles, other extra wear that, that you might have. All of these things keep changing. That is, uh, one-

Speaker 18:01 Just thinking of a cinematographer who had long hair and who has just cut it, and our producer, too, had a lot more hair and is losing it very fast. [chuckles]

Vignesh Krishnakumar 18:08 Yeah, these are very common-

Speaker 18:10 Of which

Vignesh Krishnakumar 18:10 ... set of changes that we see, right?... Uh, second, um, it has to work in all kinds of devices. Uh, you know, when we talk about process across the length and breadth of the country, essential process, right? Like opening a bank account, getting a SIM card, these things happen in every corner of the country, right? So it needs to work on every type of device that people have. It could be a 3,000 rupee smartphone, 5,000 rupee smartphone, or a high-end phone, but it needs to work on that. It needs to work on all kinds of bandwidth conditions, 2G, 3G, that's the reality.

Speaker 18:45 In fact, as you're saying it, I'm thinking that I know that DigiYatra doesn't use HyperVerge.

Vignesh Krishnakumar 18:50 [chuckles]

Speaker 18:51 I've had so many cases where DigiYatra simply is not working for me, and I have to sort of, uh, walk out of the queue, get in a... It's a very frustrating experience when you know that face match is a technology that works every time on my phone and maybe every time on a, on a HyperVerge.

Vignesh Krishnakumar 19:08 Right.

Speaker 19:08 But when you, when you experience it not working-

Vignesh Krishnakumar 19:10 Yeah

Speaker 19:10 ... it's really frustrating.

Vignesh Krishnakumar 19:12 Yeah. Exactly. So we take these problems very seriously and, uh, build specifically focused on use cases that we are solving for.

Speaker 19:22 Mm.

Vignesh Krishnakumar 19:23 So for a financial onboarding process, how can we have the best possible model in the world which can be optimized for these conditions, right? Optimized to work in these conditions really, really well. If you take the problem of liveness, or whether there is a real person standing in front of the camera, or is this a photo of another photo or a spoof, uh, right? So the way we have designed the system, um, is very different from what existed in the world or what people normally thought about. So back in 2016, '17, the, the most commonly available liveness systems used to ask the user to perform some gesture.

Speaker 20:01 Turn your head.

Vignesh Krishnakumar 20:02 Turn your head, or smile, or blink, or, uh, do one of those, uh, activities, right? Now, we were thinking about how can we imagine every person in the country has to be able to do this? Then how do you communicate these instructions across? It's going to be very hard, and literacy rate in the country is what it is, and it has to work with that. There are so many languages. How do you explain these actions very clearly to people, right? There are going to be drop-offs which are very, very significant if we expect those things to happen. So first call that we take or took was that it has to be a passive liveness system, which means the user should not need to actively participate in that, right? Uh, so you just point, and it should work.

Speaker 20:45 Yeah. Take a selfie, it's done.

Vignesh Krishnakumar 20:46 Take a selfie, it's done.

Speaker 20:47 Yeah.

Vignesh Krishnakumar 20:48 Second, the other thing that people did, okay, I cut down on gestures, but can I record a video? When I record a video, I will have, you know, a lot more frames to process. I can see the user for a longer period, and the AI models can work better, right? Um, but here, uh, again, we felt that, you know, this needs to work on all kinds of bandwidth conditions, and we might not always be able to record a video and upload a video and use that to process, and the user experience has to be really, uh, quick, right? The turnaround time has to be low. We don't want the user to drop off.

Speaker 21:24 I want to underscore that point. When you are talking of a system that processes 100 million, even a 1% change is a million people, right?

Vignesh Krishnakumar 21:30 Exactly.

Speaker 21:30 A million interactions.

Vignesh Krishnakumar 21:31 Exactly. Exactly.

Speaker 21:32 So, um, maybe somebody-

Vignesh Krishnakumar 21:34 And if you translate that to lifetime value of those 1% of users who are lost, that is a huge loss to a financial institution or any other enterprise-

Speaker 21:44 Yeah

Vignesh Krishnakumar 21:44 ... doing onboarding at scale.

Speaker 21:45 That, that's 1 million people you are saying, "No, no, you have to turn your head again." [chuckles]

Vignesh Krishnakumar 21:49 [chuckles] Absolutely.

Speaker 21:50 And they're going to-- they're going to feel bad about it.

Vignesh Krishnakumar 21:52 Yeah.

Speaker 21:53 Right.

Vignesh Krishnakumar 21:53 So this whole thing has been built as a passive, single image-based, uh, liveness, right? So, so each of these models have been built in a, in a manner where, for that use case, we are the best in the world, uh, at solving that problem.

Speaker 22:09 Nice. So what I'm understanding is that it's not a wrapper. You built your models from scratch. I've also heard, uh, Arvind, uh, from Perplexity say that: [chuckles] "How does it matter if it's a wrapper? Everything is a wrapper." And if you, if you boil it down, then human beings are wrappers of genes, and that's, that's fine. Like- The point that you're making about, uh, general purpose AI, uh, being okay for a lot of circumstances, but in a lot of special cases, you need things that are built ground up. Is that what you're saying?

Vignesh Krishnakumar 22:41 Absolutely. So our focus that way has been... And, and, um, see, we are here to build a sustainable business in the long run, right? So, um, if we know that Big Tech is going to, uh, always be, you know, on the lookout for general purpose AI models that they can launch, uh, that will always be out there. So if we have to build a sustainable business, we have to be differentiated, and we need to focus on problems which general purpose AI cannot solve. So that has, uh, largely been our focus. What we have done in KYC, in financial services is one example, but continuously our focus will be on, as the AI landscape evolves, what are those problems that general purpose AI cannot solve really well? And those are the use cases that we'll keep going after.

Speaker 23:29 Sure.

Vignesh Krishnakumar 23:29 To give an example

Vignesh Krishnakumar 23:30 of an- another company that has done this really well, um, if you think about the trajectory of, uh, Nvidia. Even before Nvidia existed, the whole computing wave was extremely hard, extremely popular, and, uh, CPUs, you know, were doing really well.

Speaker 23:46 Yeah, Nvidia started in the '90s, and Intel started in the '60s, so-

Vignesh Krishnakumar 23:49 Yeah, long back, right? And, and, uh, computing devices were hugely popular and very common in the world, and there were also high-end CPUs, which were used for high-performance computing. But Nvidia kept their focus to what are those use cases or niche problems which are not solved well by the best of CPU compute that is out there, right?... and they kept doing more and more experiments, working with research labs, understanding problem statements from various sources which are not solved sufficiently well with the best of high-performance compute that existed at that point, right? And few of those experiments became huge successes, right? What happened in GPUs for gaming, for example, that was a huge wave. Uh, Bitcoin mining at some point became a huge wave. Uh, AI, uh, deep learning, uh, and, and GPU compute for AI became a huge wave.

Speaker 24:46 Sure. You're saying that they were working on AI chips much before AI chips became a big thing?

Vignesh Krishnakumar 24:51 Exactly.

Speaker 24:52 Hmm. They were working on a bunch of very specific problems, and those specific problems became really big.

Vignesh Krishnakumar 24:57 Correct. Problems which-

Speaker 24:59 And the general-

Vignesh Krishnakumar 25:00 Which-

Speaker 25:00 Uh, general purpose CPU was not solving.

Vignesh Krishnakumar 25:02 Was not solving really.

Speaker 25:04 Nice. Uh, that's a fantastic analogy. And also, like, I like that you've taken Nvidia as an example because [chuckles] they're one of the largest companies in the world. I guess you are manifesting out there that HyperVerge will be that size and-

Vignesh Krishnakumar 25:17 Hope

Speaker 25:17 ... impactful. [chuckles] So, um, it a- on this general purpose, a special purpose thing, are there any international benchmarks that measure, uh, uh, say, general purpose face match and your face match, or, uh, general purpose liveness and your liveness? And is there, like, some way to say that, "Yes, for HyperVerge- for these purposes, HyperVerge tech is much better"?

Vignesh Krishnakumar 25:40 Yeah, absolutely. The, the Department of Homeland Security in the US conducted a benchmark recently, um, testing face recognition systems that are built for a customer selfie to ID card, uh, matching. That was the primary focus of that benchmark. So there were, uh, multiple types of tests that were conducted as part of this benchmark, and, uh, 18 of the largest, uh, and most prominent, um, identity verification companies, uh, were, uh, tested as part of this. And HyperVerge was the only company that passed all of the benchmark tests, and this was, uh, done by the Department of Homeland Security. Several other examples, uh, NIST, uh, has this global benchmark called FRVT. Uh, HyperVerge is the highest-ranked, uh, face recognition, uh, vendor in India, uh, in this global, uh, leaderboard. And amongst, uh, most other companies across the world, we are, uh, frequently present in the top 10 across the world, right? This is, uh... The, the reason I'm saying "frequently" is because it's a rolling benchmark. You can make a submission only, uh, only once in every four months, so the, the ranking keeps varying. But, uh, we are among the top 10 companies in the world.

Speaker 26:54 That's amazing. I'm just thinking that, uh, what you've said so far is that for a large bank or a telecom or anyone who needs to trust, to know its customer, and, uh, have faith and trust that he's the person he's saying he is remotely-

Vignesh Krishnakumar 27:08 Yeah

Speaker 27:08 ... um, there are three things. One is that there can be a very small room for error, um, almost zero. Because if there is error, there's, uh, poor customer experience, there's added cost because I'll have to do it again.

Vignesh Krishnakumar 27:20 Yep.

Speaker 27:20 Uh, and sometimes it's just not possible-

Vignesh Krishnakumar 27:22 Yeah

Speaker 27:22 ... right?

Vignesh Krishnakumar 27:23 And there's fraud also, right? Uh, once you open up a digital onboarding process for the benefit of, you know, millions of people out there who can be catered to, uh, for the financial product, right? Which, which is what helps with the mission of financial inclusion, uh, for the country. It also opens up the door for fraudsters to enter, right? Today, for example, uh, deepfakes are becoming a very common thing, right? You might have seen, uh, Nitin Kamath creating a deepfake video and posting it online. The deepfakes of Naran Murthy that, that was going viral. Uh, deepfakes of Trump are very popular. So all of these things, uh, we keep seeing.

Speaker 28:05 Yeah.

Vignesh Krishnakumar 28:05 But-

Speaker 28:06 I, I saw a very funny, uh, video of Trump and, uh, Elon Musk dancing. [chuckles] It's definitely not them.

Vignesh Krishnakumar 28:11 Yeah, yeah, absolutely. But the, the more scary side of these things is that when you have someone with your face-

Speaker 28:18 Hmm

Vignesh Krishnakumar 28:18 ... applying for a loan in your details and actually defaults on it, right, then there is something at stake, right?

Speaker 28:27 Yeah.

Vignesh Krishnakumar 28:27 So there's actually, uh, money that has gone out from a financial institution to an individual, and it has gone out in the name of a synthetic identity or someone else's, uh, identity or stolen credentials, and that money is lost. So-

Speaker 28:43 For sure, and for me, individually, my credit score is gone.

Vignesh Krishnakumar 28:46 Right.

Speaker 28:46 I may be, uh... You know, I, I, I may not be able to prove that I didn't do it myself.

Vignesh Krishnakumar 28:51 Exactly, yeah.

Speaker 28:51 Yeah.

Vignesh Krishnakumar 28:52 So every mistake that is done by an AI model in this process has serious consequences.

Speaker 28:58 Sure.

Vignesh Krishnakumar 28:59 Be it-

Speaker 28:59 And also, in India, we are lucky that Aadhaar exists and, uh, there's a central database which sort of verifies the identity.

Vignesh Krishnakumar 29:06 Yeah.

Speaker 29:06 In a lot of countries, that is also not there. In fact, even in India, maybe 30 years back, getting a driver's license could have been done in many ways.

Vignesh Krishnakumar 29:14 Yeah.

Speaker 29:15 Uh, so I want to ask you,

Speaker 29:18 uh, how did you get into this space? Um, what was the genesis of your journey into AI?

Vignesh Krishnakumar 29:25 Yeah. So genesis of the journey into AI goes, uh, way back, uh, before HyperVerge, uh, began as an organization. So we were a bunch of kids working on this technical club called Computer Vision Group at IIT Madras at the Centre for Innovation at IIT Madras. Back then, this technology of computer vision was very fascinating for us because even the thought that, you know, like, a computer would see and, and infer, and make decisions that a human could do was, uh, very fascinating, right? The accuracy and the capability that we had at that point of time was nowhere close to what you have right now. But, uh, it was something that, um, was very fascinating. Uh, so we started working on different kinds of, uh, hobby robotics projects, um-... in which computer vision was an important, uh, problem statement, and eventually we, we represented IITM and India in various international robotics competitions, like, um, underwater vehicles that need to navigate using vision, all-terrain vehicles, uh, flying drones that need to nav- navigate using vision. In that process, we became very good at this technology, went head-on head against MIT, CMU, Stanford, some of the best out there, and, um, learned to do really well, uh, in this area, right?

Speaker 30:50 So you were a student group called Computer Vision Group-

Vignesh Krishnakumar 30:54 Yeah

Speaker 30:54 ... in, uh, Centre for Innovation, and, uh, you would help other student groups which are participating in international competitions, uh, i- uh, get their robots to understand and infer and act better?

Vignesh Krishnakumar 31:06 Correct.

Speaker 31:06 Nice.

Vignesh Krishnakumar 31:07 Correct. What happened, um, was that over a period of time we started feeling like kids in a toy store, right? Like, you put a kid in a toy store, it plays around with, you know, all kinds of toys for an entire day also, but at the end of the day, it'll just drop all the toys and leave, right? Like nothing has changed in the world because of all of the fun that happened. So we were essentially being kids playing around with technology toys-

Speaker 31:32 Ah. [chuckles]

Vignesh Krishnakumar 31:33 ... uh, amusing ourselves with it, but nothing was changing in the world because of the work that we were doing, right? So by second year, we took this call that, uh, from now on, if we work on anything, it has to be a real-world project, um, that, you know, someone benefits from.

Speaker 31:52 Yeah.

Vignesh Krishnakumar 31:52 And that those are the problem statements that-

Speaker 31:54 And you were in your second year when you were thinking like this?

Vignesh Krishnakumar 31:57 Yes. Yeah, of course, the, the group itself had, uh, seniors, uh, before me. Uh, Kedar had started the Computer Vision Group. He was involved, uh, in this process at that point of time, so it, it was a collective decision. There were multiple people involved, and, and, you know, eventually that became the core group-

Speaker 32:16 Kedar is now the CEO.

Vignesh Krishnakumar 32:17 Now the CEO.

Speaker 32:17 Yeah.

Vignesh Krishnakumar 32:18 Eventually, that became the group that started HyperVerge, um, later. We said we will go out there and get problems from the industry and solve those problems, and that should be the primary focus for us. And the way to validate that we are creating real value for them would be that we will charge some money for that, right? It was not from the perspective of making money, uh, as students, but to get real validation that they will take us seriously and give us a problem that they really want solved.

Speaker 32:49 Right.

Vignesh Krishnakumar 32:50 So our

Vignesh Krishnakumar 32:51 first big breakthrough happened when Southern Railways had come to IIT Madras, um, with the problem statement of overhead line inspection. So electrified, uh, trains, right? So you have this overhead line that carries power. Uh, the, the engine has the, a contraption called a pantograph.

Speaker 33:10 That thing-

Vignesh Krishnakumar 33:10 Like a-

Speaker 33:11 - goes like that. Yeah

Vignesh Krishnakumar 33:12 ... spring and V-type mechanism, uh, which, um, applies pressure upwards and maintains contact with the overhead line, right? So when this contact is uniform, um, and maintained well, you have the vehicle running smoothly. If the contact is not at the center of the pantograph, imagine if it is at the side, and this is applying pressure upwards, then it starts tilting, right? And, uh, the wire can slip and snap. And, uh, you have like a twenty-five kV line that is lying on the track. In our country, people are used to walking out on the track and, and crossing lines, right? So, um, it was a, you know, real serious threat to human lives, and three such incidents had happened, uh, just in the previous one month before, uh, Southern Railways team had approached, uh, IITM with this problem statement. And this had come, uh, through to ICNSR and, um, uh, Professor Krishnan Balasubramanian sir, uh, had suggested that why not, you know, this computer vision group, uh, look into this problem statement and see, uh, if we can do something? And the reason computer vision might have been needed to solve this problem, um, or some other equivalent methodology, was because whatever electronic sensors you put on top of the engine, it might have induced current, right, and wrong reading because of it, since you have a twenty-five kV line, um, overhead. So, um, we started working on this problem and, uh, initially, you know, we had no clue what to do about this. We would go there, try-

Speaker 34:54 One second, one second. Did you say that the problem came to IIT Madras?

Vignesh Krishnakumar 34:58 Yeah.

Speaker 34:59 And a professor at IIT Madras said, "You know what? Our students in second year can just do this." [chuckles]

Vignesh Krishnakumar 35:05 No, I would say, um, and I'm very grateful that, uh, he took the call to, to expose us to these kind of industrial problems.

Speaker 35:15 He must have trusted you a lot.

Vignesh Krishnakumar 35:15 Yeah, we had done, um-

Speaker 35:18 Kedar is very persuasive, and I guess you must have done really good work-

Vignesh Krishnakumar 35:23 Yeah

Speaker 35:23 ... by ear.

Vignesh Krishnakumar 35:23 Yeah. Yeah, we, we had, um, professors guiding us through the process, um, being involved in interacting with authorities, uh, from the railways, and, and guiding us through the process. Uh, but very grateful that, you know, he took the call to, um, expose us to this, uh, real-world problem at that point.

Speaker 35:47 So then you were working with the railway to solve this issue. But the railways must have had a method before, right?

Vignesh Krishnakumar 35:53 Yeah. So the previous method that they had, um, involved, uh, running a maintenance wagon, uh, regularly from time to time. So there are three individuals who are sitting inside this wagon. Now, one person is, uh, running the maintenance wagon. There is one guy, um, who is sitting near the window, looking out and reading out a mile number. He's shouting out a mile number. So the, the system they had to identify locations along the track was to use a mile number. So this guy is shouting out a number, like mile number one, two, three, and so on.... there is a glass panel on, on top of the maintenance wagon, and, uh, there's a third guy looking up at the pantograph through the glass panel. And the pantograph has, uh, readings marked on it in pen, right? That this is-

Speaker 36:41 Human vision.

Vignesh Krishnakumar 36:42 Uh, human vision, yeah. So zero is marked at the center, and then you have, like, five, 10, 15, and so on-

Speaker 36:48 Okay

Vignesh Krishnakumar 36:48 ... and minus five, 10, minus 10, 15. And this guy is looking at the pantograph, uh, from a distance through the glass panel, and he's marking down the mass number and the reading of the point of contact, uh, of the overhead line and the pantograph from where he see, right? So, like, there is some parallel error involved and human error in the process, et cetera. So, so despite the system, um, existing, the incidents were still happening, right? So it was not, um, fully well solved, uh, uh, at that point. So every night after the regular, you know, electric vehicles and the outstation vehicles, uh, you know, are done exiting the city, post 12 o'clock, we, we used to, you know, go on our, uh, test, uh, drive, right? We, we go on top of the engine, set up a camera, come down, get into the vehicle, and then, uh, we go on this drive where we are capturing video and, and we are seeing, uh, you know, how well we are able to identify, um, if the alignment is good, if the... If you're able to detect the point of contact between the overhead line and the pantograph, the height of the overhead line, uh, from the top of the engine.

Speaker 38:03 Yeah.

Vignesh Krishnakumar 38:04 So these were the two parameters which are very critical to the health of the overhead line. And, you know, this experiment goes on till, like, 4:00 AM in the morning, and once, once this, uh, setup is done and we enter the electrified section, we can't go up and change anything, right? So you get to do one experiment, uh-

Speaker 38:21 Yeah

Vignesh Krishnakumar 38:21 ... per night. Um, so we come back in the morning, sleep, wake up, attend our classes-

Speaker 38:26 Okay [chuckles]

Vignesh Krishnakumar 38:28 [chuckles] ... uh, go in the evening, strategize about what should be the next experiment, uh, set that up, and, and the system keeps going on. So eventually, one day, things start to work, and because this is a vision system, on the screen, we can see, you know, that the outputs are showing up and things are finally starting to work, right? And, and there is this... As a, as a, you know, like a programmer and problem solver, I'm looking at it, and I feel relieved that finally things are working. Okay, whatever effort we have put over these two, three weeks, it's giving some results, so I feel happy. But this guy who used to run this maintenance, uh, wagon and, uh, and come along with us for this testing, he was looking at it, and he was like, "Sir, this is working, sir. Working, sir, working," He was so excited looking at the output. And, uh, finally that day after, after this test run, we were spending time with him to understand about his life, and, and he was telling us that, you know, during the regular day, he runs the local electric train, and, and apart from that, every time they go on this maintenance run, he has to take the maintenance wagon, and any time the line goes down, he'll have to rush, uh, do this. They'll have to get to that point, de-electrify-

Speaker 39:48 Yeah

Vignesh Krishnakumar 39:49 ... uh, do the maintenance, and then go back, right? And-

Speaker 39:51 But I'm also guessing that if something went wrong, he would be blamed.

Vignesh Krishnakumar 39:55 Yeah.

Speaker 39:55 Like, that's your fault.

Vignesh Krishnakumar 39:56 Yeah, and, and this was happening at a time when his daughter was in 10th standard, and he was feeling bad that he was not able to spend enough time with her, right? And, and

Vignesh Krishnakumar 40:06 just the thought that something like this could solve his problem, and we were nowhere close to productionizing it, right?

Speaker 40:13 Yeah.

Vignesh Krishnakumar 40:13 It was the first time it had started working. [chuckles] But just the thought that, you know, something like this could eventually solve this problem, created so much joy, uh, for him. So I tell people that if anyone else had been in my situation, they would have wanted to start a company, too.

Speaker 40:30 Can I, can I interject you there? I don't agree with you. I'm thinking of this... Uh, listen to what I, listen to what I have to say, okay? You have had the, uh, unique opportunity of getting a real-world problem in your second year. You have taken the effort, you've solved it. It's a brilliant solution. It's working. Everybody's really happy. You could easily put it in your resume [chuckles] and get a great job.

Vignesh Krishnakumar 40:48 [chuckles]

Speaker 40:48 Like, you would easy- You do a really good job in that interview, right?

Vignesh Krishnakumar 40:52 Yeah.

Speaker 40:52 So you see how the paths could have easily-

Vignesh Krishnakumar 40:54 Possibly, possibly. But in, in our trajectory, what ended up happening was that through these different stints at real-world problem solving, we could see that what, uh, you know, like 500 lines of code I would've written, uh, for... I, I could have done that for any other thing, right? For my assignments, I would have done it.

Speaker 41:15 Mm.

Vignesh Krishnakumar 41:15 So many, uh, other instances of me writing that code, that it is a ve- very small thing for me as an individual, for abilities that I've been fortunate to have had. But in another person's life, it means so much, right? It can create a huge world of impact for another person.

Speaker 41:34 That's true.

Vignesh Krishnakumar 41:34 So when you see that happening firsthand in front of you, that abilities that you're fortunate to have in another person's life creates such a drastic impact, then, then the question that comes to you is either you can use this ability to see what you can have for yourself, right?

Speaker 41:51 Yeah.

Vignesh Krishnakumar 41:51 I can have a fancy job, make money-

Speaker 41:55 That's true

Vignesh Krishnakumar 41:56 ... be rich, powerful, famous, or whatever, whatever. Recognized in one of many things. Or I can accept the fact that we are already fortunate to have this skill, and it can be used to create a lot of impact in the lives of a lot of other people-

Speaker 42:11 Right

Vignesh Krishnakumar 42:11 ... and see how can I contribute with this ability, right? So that was the thought process that all of us went through together, and that is what eventually led to HyperVerge emerging as a platform for such smart people to come together and see how they can use their abilities to contribute to society at-

Speaker 42:29 Fair enough. You're saying that if I'm a builder and I can build and I-... taken the effort to learn the tools and the tricks and the-

Vignesh Krishnakumar 42:37 Yeah

Speaker 42:37 - and, and, and I have the attitude to build, then I have a choice to make on how to use that, uh, skill.

Vignesh Krishnakumar 42:42 Exactly.

Speaker 42:43 Like, amazing. Um, uh, you also had an internship at Microsoft in the US, right, when you were at IIT.

Vignesh Krishnakumar 42:50 Yeah.

Speaker 42:50 So then you've seen that end also.

Vignesh Krishnakumar 42:52 Yeah.

Speaker 42:52 And so for you, that co- choice is a very conscious choice then.

Vignesh Krishnakumar 42:55 Yeah.

Speaker 42:56 It's amazing.

Vignesh Krishnakumar 42:56 Yeah, it was a very conscious choice. And, and through these internships, you also get the, um, understanding that this ability is valuable, right? That whatever, uh, we have learned... See, till, till that point of time, almost everything was about being at the right place at the right time. Uh, and there is nothing that I can take credit for. A lot of things happened.

Speaker 43:22 I mean, you can take credit for crossing the [laughs] and getting a good rack.

Vignesh Krishnakumar 43:26 Well, I, I mean, anything could have happened, right? I was fortunate that I was born in a family where they could afford to send me to, to the right school. I got educated. I had a good set of teachers, good mentors, upskill myself for the people-

Speaker 43:42 That is first

Vignesh Krishnakumar 43:43 ... who exposed me through that.

Speaker 43:44 Yeah, that's-

Vignesh Krishnakumar 43:44 Uh, got into IITM, again, teachers, uh, the Centre for Innovation Ecosystem, CVG as a club. All of these things happened, so the abilities that I had at that point of time, um, I should say I was fortunate to have that. And when, when I worked for these internships, I can see that these abilities are useful in the world, right? That they're actually valuable in solving problems for the world, and you understand and appreciate the value of it.

Speaker 44:16 Yeah.

Vignesh Krishnakumar 44:16 So that led me, uh, to believe even better that I should use these abilities for the well-being of people to solve problems that can impact, you know, people at scale, and that would be a great use of this ability that we have.

Speaker 44:33 Amazing.

Vignesh Krishnakumar 44:34 If everything fails, there is always JEE coaching, [chuckles] JEE coaching as a backup job. That was our, uh, fail-safe at that point of time.

Speaker 44:43 That is amazing. You were saying that, uh, a lot of luck was involved in you getting here, and a lot of things that other people have done-

Vignesh Krishnakumar 44:51 Yeah

Speaker 44:51 ... are pouring their efforts into you.

Vignesh Krishnakumar 44:53 Absolutely.

Speaker 44:54 Uh, I, I'm also thinking that while this is a choice that you made, that you were... Of course, you were a founding member of HyperVerge. There were other founders. There were also a lot of other employees who joined you-

Vignesh Krishnakumar 45:04 Yeah

Speaker 45:04 ... from the same ecosystem. So it's not an individual, it's now almost 15, 20 people choosing the same thing together.

Vignesh Krishnakumar 45:09 Yeah.

Speaker 45:09 Uh, giving up lucrative offers and like, of course, I mean, in, in hindsight, great decision because a billion people use your technology,

Speaker 45:18 so...

Vignesh Krishnakumar 45:18 Yeah.

Speaker 45:18 Tell us a little bit about your IIT life. Um, how were you in class? How were you...? D- did you have a nickname? Which hostel were you in? Um, were you a POR, or I don't know?

Vignesh Krishnakumar 45:31 So I was in, uh, uh, Saras, uh, Saraswati hostel, and, um, [clears throat] um, went to, uh, CS dual degree between 2010 and 2014. Um, and, and I think, uh, I was, I was very fortunate that this particular, uh, branch also happened, uh, for me. And, and a lot of things that I'm doing today are, uh, aligned with what I, uh, spend time on from, uh, my classes as well, right? For, for a lot of people, it, it may not be that way post, uh, post, uh, insti days, that they do something.

Speaker 46:14 That's like... Yeah, because you go to class, and maybe you have done really well in class, but suddenly you realize that the thing that you're very good at is actually not extremely valuable at that point of time. The companies don't want to pay you for doing that.

Vignesh Krishnakumar 46:27 Yeah.

Speaker 46:27 Okay, so you end up doing something else.

Vignesh Krishnakumar 46:29 Yeah, so I, I was very fortunate in that way, that, um, the, the time that I spent in my classes also became a, a force multiplier of, uh, what I was doing otherwise, right? I had CFI and the computer vision group and, and the projects and, and eventually, uh, HyperVerge. And, and we started HyperVerge while I was still in college. So there was a lot of, uh, support from, uh, professors, uh, in the department. Professor Anurag Mittal, um, um, uh, I've done, like, multiple computer vision courses, uh, with him. He was very instrumental in many of these discussions on, uh, computer vision and, and machine learning. Professor, uh, Ravindran, uh, again, did multiple, uh, machine learning and related courses, uh, with him.

Speaker 47:18 Professor Ravindran is now the head of the new department of-

Vignesh Krishnakumar 47:21 Yeah

Speaker 47:21 ... Bhawani School of Data Science and AI.

Vignesh Krishnakumar 47:24 Yeah, very, very happy for that, and very happy that, uh, the AI ecosystem in, in IIT Madras, uh, is evolving and, and doing really well, uh-

Speaker 47:33 And, and that it would be amiss to have skipped the fact that, uh, one of, uh, Professor, uh, Ravindran's students is one of the co-founders of Perplexity. He's the same course as you, I guess. Did you know him on campus?

Vignesh Krishnakumar 47:45 Yeah, yeah, I, I knew him in campus. Um, he went to the same, uh, JEE coaching center as well, so, uh, been, uh, interacting with him, uh, from, uh, before the time he, he joined IITM, and, uh, have had a few interactions through, through the time here.

Speaker 48:02 Nice.

Vignesh Krishnakumar 48:02 Knew him from there.

Speaker 48:03 Wonderful. Um, uh, you were talking about how you were in Saras, and, um, were you, uh... I'm, I'm just intrigued with this idea that, uh, students at IITM start companies. Um, and, uh, to us in IIT, it feels like, yeah, this is what we do, like, we start sometimes it works, sometimes it scales, sometimes it doesn't. From somebody looking from outside, it must be feeling like a totally different world, right? Like, did you, as a student, ever think that you would-

Speaker 48:36 ... be running a startup in your second year [chuckles] or third year?

Vignesh Krishnakumar 48:39 No, absolutely not. Abso- absolutely not. And, and the process of making this into a startup was also not, not easy because, um, all of us come from, um, all the people who were part of this computer vision group back then, which became HyperVerge, um, eventually. We, we came from, uh, very normal middle-class families.

Speaker 49:02 I guess in telling your parents that-

Vignesh Krishnakumar 49:04 Yeah

Speaker 49:05 ... I'm not going to sit for Microsoft, [chuckles] what have you done?

Vignesh Krishnakumar 49:08 So-

Speaker 49:09 Call

Vignesh Krishnakumar 49:10 ... parents who have been like government employees, bank employees, et cetera, and, and people who spend their entire lifetime working for an organization like that, right? And, and, um, telling them that I'm not going to sit for placements was not an easy thing at all. So I, I told my mom, uh, about this, and, um, you know, uh, she, she did not understand this concept of not sitting for placement very clearly. And then on the day of placement, someone, uh, in her office happened to tell her that, "You know, today is the placement day at IIT Madras." She calls me like, "Go sit for placements today. Like, do-- what is happening for your interview?" I told her, "I didn't register for placements. Like, [chuckles] what are you telling me?" Uh, so, uh, and then that was already like a year into starting HyperVerge, right? So it is not easy for, uh, parents to understand that, especially when they've not seen enough people, um, in, in our-

Speaker 50:10 Yeah

Vignesh Krishnakumar 50:10 ... communities, in our-

Speaker 50:11 Yeah

Vignesh Krishnakumar 50:11 ... social circles, who, who do that. Um, and when everyone else from your class is not doing that, and one of something else, they feel, you know, worried about, "Is this the right choice for my kid?" Right? And it comes from a place of very, very good intention and, and wishing the best for, for their child, but without a lot of validation, it becomes very hard. So what we decided to do, uh, back

Vignesh Krishnakumar 50:38 then, at, at the end of my, uh, fourth year, uh, Kedar and I decided to, uh, travel to the US, and, uh, we thought we'll do like a Kickstarter campaign of sorts and raise some money from, uh-

Speaker 50:52 Hmm. So it'll de-risk if you have, say, money in the bank. Yeah.

Vignesh Krishnakumar 50:54 Tech-savvy people from the Valley who put in some money, and we will use that as validation to take to the parents of all of these people that, "You know, your kid need not go to one of these places, and we are up to something good, and-

Speaker 51:08 Yeah

Vignesh Krishnakumar 51:08 ... you know, people in the Valley who understand this technology are, uh, putting in money," right? So-

Speaker 51:14 I mean, a lot of times funding works like that. If somebody's giving you so much money, [chuckles] you must have something going good.

Vignesh Krishnakumar 51:19 Yeah. So we were one of the early teams to get into deep learning, um, uh, and, uh, back then, uh, 2013, '14 in In- India, it was not, um, very common or popular, and, and, uh, we thought... I mean, it may or may not have been factually true, but at least we thought people in the Valley are the people who are going to understand this technology, and, uh, we decided to go there, right? And at that time, uh, we had built, uh, one of the first deep-learning-based scene recognition engines, um, in the world, and amongst the, uh, early set of teams to-

Speaker 51:59 Scene recognition?

Vignesh Krishnakumar 52:00 Yeah. So you give it any photo, it'll identify what is the context in this photograph. Is it a, is it a beach or a café or a forest or a meeting room or an office space or whatever it is, right?

Speaker 52:13 This is much before you decided to move into KYC?

Vignesh Krishnakumar 52:16 Yeah, yeah.

Speaker 52:17 Yeah.

Vignesh Krishnakumar 52:17 Much, much before. This was, um, 2013, '14, that, that I'm talking about. Um, so, uh, today, these kind of image recognition APIs are very common. You give it an image, it'll predict what is the context, uh, in it, right? Uh, but back then, um, people had not seen anything like this, and, and deep learning as a buzzword was, was becoming, like, very, very popular. Um, Andrew Ang had shifted to the Google Brain team, and there was this project, uh, about detecting cats from YouTube videos, right? And, uh, when, when Google sneezes, the whole world catches cold. So, uh, deep learning as a, as a topic was going viral, but no one knew what was a real use case for it, right? What will you do detecting cats from YouTube videos? So using this scene recognition and face recognition, we said we will build a consumer app which will sort through the photos in your gallery, and you can search for things like my photos with a friend at the beach, right? And pull up the memories that you wanted to revisit. Um, and we thought we'll do a Kickstarter campaign of sorts for, for, for this app and, and the technology behind it, and raise some money, uh, from the US, and use that as validation. But when we went there and we had put together a demo where on one side, uh, you had Google Image search, and on the other side, our, um, you know, image recognition engine. And we u- we used to tell people, like, "You search for anything," and we'll just crawl and ask them to pick a photo, and we'll just, like, drag and drop that here, and instantly, in less than a second, it'll predict what that image is. And if it is not any of the scene categories, it'll also learn to tell that it is not one of these, right? It is not one of, you know, the, the, the scene categories that we expect it to predict. So, so people went crazy playing around this, right? It was the first time they had, like, direct access to like a deep learning tool that they could, you know, play around with and give it some real samples and see it predict their stuff. So, uh-

Speaker 54:19 This is a very real problem. You have so many photos on your phone, you always wonder, "Can I just, you know, pull up all the beach photos or all the, uh, screen...? At least can I pull up all the screenshots and- [chuckles] -

Vignesh Krishnakumar 54:30 ... delete, delete them?" Yeah. Yeah, common problems that, that we used to hear, uh, back then, and, and this model, you know, was capable of doing all of that.... um, but, uh, when we met, um, people in the Bay Area, alumni of IITM, uh, entrepreneurs who had built companies before, um, uh, VCs, investors, people told us that, "See, you can do your Kickstarter campaign and all of that, uh, build your consumer app, launch it if you wish, but the, the technology behind this is way more powerful than what you're thinking of-

Speaker 55:06 Okay

Vignesh Krishnakumar 55:07 - today, right? And you will end up solving really powerful problems for the industry if you, if you take it there. And as a consequence of that, you'll end up spending a lot more money also, because these are larger problems to solve, and you'll need a team and, you know, uh, proper mechanisms to go after that kind of a market. So raise a lot more money than you're thinking of today. Ditch your Kickstarter campaign, go talk to VCs, and raise-

Speaker 55:34 Like, a million dollars or-

Vignesh Krishnakumar 55:35 Yeah.

Speaker 55:35 Maybe five or so.

Vignesh Krishnakumar 55:36 So that is what people told us, and that is exactly [chuckles] what we did in that trip, because we had, you know, no idea, uh, what to do. It was, you know, amongst the first few times we were being in the US, we didn't know whom to meet. Again, uh, the entire IIT Madras-

Speaker 55:50 How, why- wait a second. Hold on. Who paid for this? Like, you were still students, right? You were in fifth year.

Vignesh Krishnakumar 55:55 Yeah, yeah. So I, uh, we had incubated ourselves as part of the IITM incubation cell-

Speaker 56:02 Yeah

Vignesh Krishnakumar 56:02 ... uh, here, um, at IITM.

Speaker 56:04 You guys were being one of the early incubatees.

Vignesh Krishnakumar 56:06 Early incubatees, uh, in the IITM research park, uh, ecosystem under IITM IC. And, uh, we had gotten a initial funding round, uh, through IITM IC, uh, book, and that is what, uh, helped, uh, fund this travel initially and, and the process that we did in the US at that point of time. And, and not just that, um, uh, you know, tremendous amount of support, uh, through that entire process. Uh, Seshan Raman sir, who, whom we call fondly as the granddad of, uh, HyperVerge. Uh, again, uh, IITM, um, alum, who, who, uh, took us, uh, to places and, and back then, you know, we would not have, uh-

Speaker 56:52 He's done this to so many people.

Vignesh Krishnakumar 56:54 So many, so many entrepreneurs, and we have w- personally recommended many such juniors, upcoming entrepreneurs, to go meet him. Like, you know, first thing that you do when you land up in Bay Area, go meet Seshan sir, right? Everything else-

Speaker 57:08 Yeah

Vignesh Krishnakumar 57:08 ... you know, you can do after that.

Speaker 57:09 Right.

Vignesh Krishnakumar 57:09 Yeah. Um, and-

Speaker 57:10 I think, uh, people outside the network underestimate how much support alumni give to-

Vignesh Krishnakumar 57:17 Yeah

Speaker 57:18 ... uh, their juniors. Because simply, I don't even know why. They'll just say, "You know what? I'll connect you," or, "I'll help you," or...

Vignesh Krishnakumar 57:24 Yeah, and there's been an amazing, uh, pass-it-forward, uh-

Speaker 57:28 Yeah

Vignesh Krishnakumar 57:28 ... culture, right? So they help, and, and there is just absolutely no expectation. And, and for the gratitude that you have for all the support that you received, you want to, you know, pass it on to other people who are, who are on that path, uh, after you. So, so that culture has been tremendous, and you've seen many, many such, um, seniors and, and alums, uh, do that for us. And, and that is what has been propelling this ecosystem, uh, forward.

Speaker 57:56 Amazing. So you built that app. I, I remember it was called... Was it called Silver?

Vignesh Krishnakumar 58:00 Silver. Yeah, it was called Silver at, at one point. [chuckles]

Speaker 58:04 And, and something happened, right, like, to Silver?

Vignesh Krishnakumar 58:05 Yeah, yeah. So 2015, '16, uh, Google Photos launched, um, and, and they did the two things that, uh, no startup could compete against. They made unlimited photo storage free-

Speaker 58:19 Yeah

Vignesh Krishnakumar 58:19 ... um, and they became the default gallery app on most Android phones, right? Both, both of which only, uh, Google and an Apple could have probably done.

Speaker 58:28 Yeah, and, uh, you know, the threat, this is back in 2015, you're always worried. Everybody would ask you, "What if Google does the same thing?"

Vignesh Krishnakumar 58:35 Yeah.

Speaker 58:35 And today, every, [chuckles] every AI startup-

Vignesh Krishnakumar 58:38 Yeah

Speaker 58:38 ... asks the same question: "What if ChatGPT or-

Vignesh Krishnakumar 58:42 Yeah

Speaker 58:42 ... Perplexity does the same thing?"

Vignesh Krishnakumar 58:43 Absolutely.

Speaker 58:44 Overnight, you're wiped off.

Vignesh Krishnakumar 58:45 Yeah.

Speaker 58:45 So that did happen to you.

Vignesh Krishnakumar 58:46 That-

Speaker 58:46 You were overnight-

Vignesh Krishnakumar 58:47 Yeah, yeah.

Speaker 58:48 Yeah.

Vignesh Krishnakumar 58:48 That absolutely happened. And, um, fortunately for us, um, we were, uh, still in college. I mean, there was a bunch of seniors, juniors, uh, who were doing this together. I had just graduated. There were more people who were still in college, were coming out, yet to come out at, at that point, right? So our, our burn was, uh, super low. Uh, we could, um, regroup as a team, think about what we are up to. That is on one side. On the other side, through the journey, uh, fortunately we had built very, very good fundamental, uh, models at, at that point, right? The deep learning models that we had built for, uh, be it scene recognition, be it face recognition, were amongst the best in the world, even at that point of time. So the core technology was really powerful. That was on, on the other side. And we needed to figure out what is it that we wanted to do from here onwards-

Speaker 59:43 Isn't it?

Vignesh Krishnakumar 59:43 ... right? That is when, uh, we came together as a team, and deep learning was very hot at that point, and even hotter than, uh, when we had raised funds. And these big

Vignesh Krishnakumar 59:53 tech companies were acquiring startups left, right, and center, acquiring, uh, people, you know, whole research labs, and all of these things were, uh, happening common. So HyperVerge itself had, like, three acquisition offers from, um, larger companies, and we were, uh, 12 or 13 people at that point of time.

Speaker 1:00:12 That's fair. Even if you are a company where, uh, your main product has suddenly become irrelevant, if you have built an underlying technology, you have a great team-

Vignesh Krishnakumar 1:00:22 Yeah

Speaker 1:00:22 ... then you are still very, uh, attractive for a acquisition.

Vignesh Krishnakumar 1:00:25 Yeah, yeah. And, uh, double-digit million-dollar acquisition would have meant financially for, for the work that we had done, we would have made a good exit, and it is significant enough that we don't have to work [chuckles] ever again for, uh, money, right? So, so that was on one side, but, uh, the team really came together, and all of us shut ourselves up in a room-... and we said, "You know, we are going to seriously think about why we came together. What is it that we have done? What is it that we really want to stand for," right? Because if we take this acquisition offer, then post that, we all go our own ways, and it's difficult to put together the team back, um, right?

Speaker 1:01:05 This is the famous IKP-

Vignesh Krishnakumar 1:01:07 Yeah

Speaker 1:01:08 -uh, pantry meeting. [chuckles]

Vignesh Krishnakumar 1:01:09 IKP pantry meeting.

Speaker 1:01:10 Correct. Correct, Jesse. For those of you who don't know, everybody has heard... In our circles, everybody has heard this me- uh, story, where HyperVerge product was no longer very relevant. They sort of closed themselves in this, uh, sort of a pantry room with a meeting room adjacent in, uh, IKP in Koramangala.

Vignesh Krishnakumar 1:01:27 Yeah.

Speaker 1:01:28 And, uh, they spent two days trying to figure out what to do next.

Vignesh Krishnakumar 1:01:31 Correct.

Speaker 1:01:32 Right.

Vignesh Krishnakumar 1:01:32 Yeah. The first thing that we understood from our reflection was that, um, see, we started off HyperVerge as a platform for people to come together, use their abilities to contribute to solving problems for people, right? And, and helping society at large in a very meaningful way. But what ended up happening in the process was the, the first segment that we had easy access to was other college students, and, uh, uh, they ended up having a lot of clutter in their gallery, and, um, it was, uh, an easy thing to get into. But, um, we, we called this category of problems as perception problems, right? So imagine like, you know, we wear, uh-- we buy a new shirt and wear it, we, we can feel happy about it for maybe five minutes or 30 minutes or whatever, but it's not like the fundamental quality of our life changes for the better, right? So today you can have your photos sorted, you can share it better, you can feel nice about it, but it's not like the fundamental quality of your life changes because of it. So this category was what we called perception problems, and what we really wanted to do was focus on fundamental problems, where whether you like it, you don't like it, you ask for it, don't ask for it, if, uh, an improvement happens in that area, the quality of your life automatically changes for the better.

Speaker 1:03:00 Sure.

Vignesh Krishnakumar 1:03:00 For example, think of education, right? As a kid, we might have been crying about going to school, but twenty years later, the life of a person who had access to education versus someone who did not have access to education would have diverged so much, right? So education, healthcare, uh, financial inclusion, poverty, climate change, some of these are more fundamental in, in that sense.

Speaker 1:03:24 I understand what you're saying. So you're saying that among all the things you could solve, there's a set of perception problems.

Vignesh Krishnakumar 1:03:30 Yeah.

Speaker 1:03:30 What you're calling as percept... By the way, not to demean them.

Vignesh Krishnakumar 1:03:33 Yeah.

Speaker 1:03:33 Like, I'm very happy that Google Photos gives me memories from five years back-

Vignesh Krishnakumar 1:03:36 Yeah

Speaker 1:03:36 ... or ten years back.

Vignesh Krishnakumar 1:03:37 Yeah.

Speaker 1:03:37 Um, uh, and, uh, there are other problems which maybe cause upward mobility, uh, somebody gets, uh, a better access to education, somebody's cured of illnesses or whatever. Uplifts the entire country out of poverty. Okay.

Vignesh Krishnakumar 1:03:53 So that is what we called as, uh, fundamental problems, and, and we felt that, for us, that is what we are most, uh, passionate about, uh, solving, right? And, and we decided that we will not take, uh, these acquisition offers, and we will build a company focused on solving these fundamental problems, right?

Speaker 1:04:16 Okay.

Vignesh Krishnakumar 1:04:17 So we went to our investors and told them [chuckles] that that was the tricky part, that, uh, "You've put in money, but we are not going to take this exit, uh, and instead we are going to work on-

Speaker 1:04:29 Oh, right! If you're taking the acquisition, the early investors would have also gotten-

Vignesh Krishnakumar 1:04:32 Investors would have also gotten an exit. Um, "But we are going to work on these fundamental problems." And, and they told us, uh, you know, one very important, uh, thing, that, "See, these problems, um, that we are talking about have existed for ages." Um, and today there are, like, three hundred to four hundred million people in India alone who are below sustainable income levels.

Speaker 1:04:58 Yeah.

Vignesh Krishnakumar 1:04:58 Right? These problems have existed for ages, and it is going to take many decades of effort to make an impact in that area.

Speaker 1:05:07 Right.

Vignesh Krishnakumar 1:05:08 Um, think of someone like an Elon Musk, who says, "I want to, you know, convert a significant percentage of the population to driving on renewable energy," right? It is not going to happen overnight. It is going to take many decades of effort for a significant portion to, you know, make the shift.

Speaker 1:05:23 Yeah.

Vignesh Krishnakumar 1:05:23 So, um, any of these fundamental changes is going to happen over a long period of time, and as a startup, uh, by definition, you don't have time on your side, right? So, so first figure out how you're going to make that happen. Around

Vignesh Krishnakumar 1:05:39 that time, we started meeting a lot of, you know, other, uh, people, other entrepreneurs, and, and went and told them, like: This is what we want to do, but we have no clue how to go about it, right? And we're very fortunate in that we met Sridhar Vembu at that point. Sridhar is the founder and CEO of, uh, Zoho.

Speaker 1:05:57 Also an alumnus of-

Vignesh Krishnakumar 1:05:57 Also an alumnus of IIT Madras. We had, um, and we made-

Speaker 1:06:01 I think his co-founder is also an alumnus.

Vignesh Krishnakumar 1:06:02 Also, yeah. So Sridhar told us that, um, if you want to contribute by solving any of these fundamental problems, um, then you have to build an organization that stands the test of time, uh, because it's going to need that much time to create a difference in any of these areas, right? And to, um, earn time on your side, uh, you need to be-- you need to have the economic ability also. You need to be financially sustainable. You need to be able to stand on your own feet for a long enough period of time and be able to solve any of these problems, right? Third thing, you need competence on your side. You need, uh, people with really strong capabilities.

Speaker 1:06:49 Right.

Vignesh Krishnakumar 1:06:50 And these problems are unsolved for a reason, so if you-... uh, have really good competence, then, uh, you can go at solving, uh, any of these problems, right? So build an organization that has these three ingredients: that has time on its side, that has economic ability, that has, uh, competence, right? And to do that, um, first make your own money, right? That is what helps you earn these three, right? So, uh, he said-

Speaker 1:07:21 A great point. Like, if I look at Zoho now, and, uh, Zoho set up office in Tenkasi and so on, uh, and, uh, he speaks very strongly about, uh, building offices in rural India, and, yeah, it's sort of like anti-urbanization or re-urbanization.

Vignesh Krishnakumar 1:07:36 Okay.

Speaker 1:07:37 That is something he can do only at the scale of Zoho. He's Zoho, so he's able to do it. Uh, you and I can't. Maybe HaverVerge can, but, uh, definitely I can't.

Vignesh Krishnakumar 1:07:45 Yeah.

Speaker 1:07:45 So that-

Vignesh Krishnakumar 1:07:46 So give examples of how, um, um, if you build an organization for the long run-

Speaker 1:07:54 Mm

Vignesh Krishnakumar 1:07:54 ... that is, uh, that has these three ingredients on their side, and they have the heart in the right place, then through every touch point that they come in contact with people, they will only figure out a way to contribute, right? So a few examples from Zoho story that you shared. One, Zoho has set up this thing called Zoho University, where-

Speaker 1:08:13 Right

Vignesh Krishnakumar 1:08:13 ... they take people from villages, remote rural areas, people who have even just finished 12th standard, um, enroll them in Zoho University, and help train them in essential tech skills, right, that an organization like Zoho itself will need. And post that course, they are absorbed in Zoho, right? So imagine, um, a, a, a person who, who hasn't, uh, you know, used a laptop in their life at all or, you know, first generation-

Speaker 1:08:44 Yeah

Vignesh Krishnakumar 1:08:45 ... um, person who's getting exposed to technical skills, uh, them upskilling themselves, uh, them getting a job, well-paying job in the tech industry. Uh, it transforms the situation, one, in that family, um, because not just do they earn salaries which help them sustain the family, two, they get access to health insurance, um, which takes care of the, the family's health situation as a right. Uh, in a lot of these families-

Speaker 1:09:14 Absolutely. I mean, contrast that to our stories, where we studied in Chennai in top schools, uh, went to top coaching centers, came to IIT, where we were taken care by this [chuckles] amazing, beautiful CFI IIT-

Vignesh Krishnakumar 1:09:26 Yeah

Speaker 1:09:26 ... system. So yeah, totally understand what you're saying.

Vignesh Krishnakumar 1:09:29 Yeah. So that, that happens, and second, um, they will always have people around them, right? Like some brother, sister, cousins, friends, other people from the ecosystem, and they, uh, become role models for-

Speaker 1:09:44 Right

Vignesh Krishnakumar 1:09:44 ... this entire community, right? So wherever, uh, this happened, um, we saw examples of other people, you know, in, in the town, in the village, looking up to this individual, also getting enrolled in Zoho University, also in upskilling themselves, also getting jobs, um, uh, in the tech industry. And, and this, this was transforming, you know, community after community.

Speaker 1:10:09 Yeah.

Vignesh Krishnakumar 1:10:09 And this was very powerful in solving the problem of bringing these 300, 400 million people to sustainable income levels, right? The other example that he gave was when Zoho was doing really well, uh, and at its peak they were in Chennai, they could have said, "You know, I will bring all people to the headquarters in Chennai," right? "I'll recruit-

Speaker 1:10:32 Right

Vignesh Krishnakumar 1:10:32 ... talent from across the country and bring them here." But, um, when that happens, then you don't have, uh... When the smart people from these local villages and, and towns start migrating to the cities, then, uh, there are not enough, you know, uh, smart people left back in those communities who can work towards the upliftment of those societies.

Speaker 1:10:55 Maybe we can use the word skilled people, because I'm sure the others are smart enough, but-

Vignesh Krishnakumar 1:10:59 Yeah

Speaker 1:10:59 ... maybe in a different way.

Vignesh Krishnakumar 1:11:01 Probably, yeah. Um-

Speaker 1:11:02 But I get what you're saying, is-

Vignesh Krishnakumar 1:11:03 Skills and exposure, uh-

Speaker 1:11:05 The highest skilled person who can draw the highest salary in a, in a town moves away-

Vignesh Krishnakumar 1:11:10 Mm

Speaker 1:11:10 ... uh, versus if he had stayed behind, he would have paid local taxes, he would have maybe built a house, given some employment.

Vignesh Krishnakumar 1:11:16 Yeah.

Speaker 1:11:17 The area sort of benefits.

Vignesh Krishnakumar 1:11:19 Exactly. So what they ended up doing was creating smaller centers of Zoho campuses-

Speaker 1:11:26 Yeah

Vignesh Krishnakumar 1:11:26 ... in various other parts of the country, rather than bringing people to one central hub in the city, right? So when these things started happening, then, um, not only do people who take up jobs in Zoho stay back in, in the zone commun- in, in their own communities, uh, but the entire ecosystem around the Zoho campus also starts evolving, right? People get, uh, jobs as, um, you know, uh-

Speaker 1:11:54 Support staff.

Vignesh Krishnakumar 1:11:55 Support staff, uh-

Speaker 1:11:57 Right

Vignesh Krishnakumar 1:11:57 ... educational hubs start-

Speaker 1:11:58 Yeah

Vignesh Krishnakumar 1:11:58 ... coming up around the-

Speaker 1:11:59 Yeah

Vignesh Krishnakumar 1:12:00 ... schools start developing-

Speaker 1:12:01 Yeah

Vignesh Krishnakumar 1:12:01 ... hospitals start developing, housing societies start coming up. People get jobs in these house- housing societies. People get jobs in these hospitals, in education institutions. Uh, imagine stationery and other equipment that needs to be provided, transport starts improving.

Speaker 1:12:17 Yeah.

Vignesh Krishnakumar 1:12:17 All of these things start developing around that ecosystem.

Speaker 1:12:20 Absolutely.

Vignesh Krishnakumar 1:12:20 So he gave, uh, examples, multiple such examples of how by virtue of Zoho being, you know, uh, an organization that stood the test of time and doing really well as an economic engine, they were also able to power these contribution engines and actually, um, help the society at large.

Speaker 1:12:41 Understood. So from this idea that you have that, "I want to solve fundamental problems, I want to have a deeper impact in the world"-

Vignesh Krishnakumar 1:12:47 Yeah

Speaker 1:12:48 ... um, Sridhar is saying, "Hello, first take care of yourself. Have a way to earn enough revenue."

Vignesh Krishnakumar 1:12:54 Yeah.

Speaker 1:12:54 Uh, and then maybe the world gives you that permission to do these things.

Vignesh Krishnakumar 1:12:58 Correct. Correct. Our key takeaway was about building very strong, sustainable, long-term economic engines which have time, economic ability, and competence on their side. Uh, but built with-... the right awareness that we will use all of the economic ability that is coming out of this for the purpose of contribution rather than accumulation. So strong economic engines, that power contribution engines, became the vision that, uh, we started taking forward from that point.

Speaker 1:13:30 Right. So when you walked out of that room, that is what the 15 or 20 of you thought, that this is the attitude we'll have going forward?

Vignesh Krishnakumar 1:13:38 Yeah, that's right. And the action plan from therein became simple. To earn that economic ability, we had to make our own revenues, so we opened up the tech stack that we had built to enterprises, uh, started operating as a B2B business, and started generating real revenues from there.

Speaker 1:13:58 Nice. Eventually, you moved into KYC, and we know the story, and, uh, 1 billion people use. So great, fantastic. I h- I get this, um, uh, um... actually, we'll come back to contribution a bit later. Um,

Speaker 1:14:10 when you look around you now, there is a huge number of, uh, AI startups that are coming up. There's a lot of momentum. Uh, did you see this coming? Did you-- Because you've been doing it since 2012, '13, did you expect this sort of, like, mass AI hysteria to come in?

Vignesh Krishnakumar 1:14:30 Yeah. We, we felt that, uh, AI was certainly an area that, um, was going to explode because the use cases where, uh, society at large could benefit from, um, uh, AI solving those problems really well, right? So capabilities that had, uh, opened up, uh, post deep learning, uh, even, uh, back then, we felt was going to be very, very powerful in solving some of these use cases, and that it was going to grow. What, um, was interesting was the pace at which, uh, the space has evolved, uh, since then, and, uh, that was certainly something that, um, uh, we might not have predicted, but in a positive way, a lot of things have happened. Um, it has also opened up a lot of, um, other gaps in terms of, um, uh, what it enables fraudsters to do.

Speaker 1:15:33 Yeah, sure. Deepfakes, for example-

Vignesh Krishnakumar 1:15:34 Yeah

Speaker 1:15:35 ... has also opened up as a-

Vignesh Krishnakumar 1:15:36 Exactly. What security risks, um, are posed because of it, um, explainability, uh-

Speaker 1:15:45 Yeah

Vignesh Krishnakumar 1:15:45 ... privacy. Some of these, uh, other topics have become, um, more important and relevant today, uh, um, in, um, in line with the goal in itself.

Speaker 1:15:59 In, in KYC itself-

Vignesh Krishnakumar 1:16:01 Yeah

Speaker 1:16:01 ... five years back, maybe the biggest concern was, "Am I identifying the right person with a fraud ID card?" But today, it's like, "The video that's been submitted, is it a AI-generated video," right?

Vignesh Krishnakumar 1:16:13 Yeah.

Speaker 1:16:14 Yeah.

Vignesh Krishnakumar 1:16:14 Very real problem.

Speaker 1:16:15 Very real problem. Mm. So you're saying that, um, with the access you had, you could see that it's going to come, but the pace at which it's come has surprised you?

Vignesh Krishnakumar 1:16:25 Yep.

Speaker 1:16:25 Is it, uh, is it- s- has it added pressure on you? Do you, do you feel like you need to grow faster or, I don't know, do more things or add new products? Is that happening?

Vignesh Krishnakumar 1:16:36 See, in the, in the, in the lifetime of building an organization, market evolution is a factor that we always have to take into picture, and we have to be very, very dynamic to these changes that happen, uh, in the market. Be it, um, the technology landscape evolving, be it, um, the regulatory landscape evolving, be it, uh, changes with our customers, competitive landscape. All of these things, um, are very dynamic and constantly keep changing, right? So it is very important to build an organization that, um, can adapt to these changes, uh, very rapidly and do what is needed. In our case, specifically, right from the early days, our, um, focus was to do things that will not get disrupted by general purpose AI. Was-- It was always to, um, solve problems that are very niche and use-case specific, and in that use case, uh, even if we are going head-to-head against the biggest tech companies in the world, um, it had to stand out. And a difference between, uh, 95% versus a 99.9% in that area should have a very serious consequence, right?

Speaker 1:18:00 Right.

Vignesh Krishnakumar 1:18:01 So those are the problems that, um, we are going after. Uh, so every subsequent product that we are working on, um, future innovation that is happening is al- also targeted towards that category of problems.

Speaker 1:18:15 That is interesting. You are saying... Uh, let me see if I got this right. You're saying that from an industry point of view, there are some problems that the industry is throwing at people.

Vignesh Krishnakumar 1:18:25 Yeah.

Speaker 1:18:25 On the other side, a lot of these AI companies, which are, uh, AI products which are coming out-

Vignesh Krishnakumar 1:18:29 Yeah

Speaker 1:18:30 ... are not really responding to the product, but responding to some new technology. So maybe some of them may not stick for a while, or by the time the problem is recognized, um, there'll be much more nuance, uh, that you'll have to work for. Is that a right interpretation?

Vignesh Krishnakumar 1:18:43 Or, or they have to build moats in other ways. Uh, technology itself might not, um, create a sustainable advantage.

Speaker 1:18:52 Mm.

Vignesh Krishnakumar 1:18:52 Um, technology, uh, leveraging some of these advancements in AI can help them, uh, solve a problem right now, and get a foot in the door, um, uh, with customers, um, and build a business with them. Uh, but through their engagement with these customers, they might have to figure out some other moat as to why they will not get replaced. Um-... when, uh, technology- I mean, everyone else also has the same technology and is able to meet the same requirement in the market. Uh, but every business fundamentally will have to figure out what is their moat for, uh, not getting replaced.

Speaker 1:19:32 Yeah, fair enough. So you have- you've developed your technology from scratch, you know it much better, you've worked with customers for five, six, seven years, so, uh, it's not like you are very threatened by new AI companies coming up and doing similar things. That's interesting. I

Speaker 1:19:48 also feel like when you are giving this answer, um, it's a very CTO answer, right? [laughing] But you started as a student club. So how have you undergone this personal journey from, you know, an engineer who's very fascinated that his 500 lines of code has made this wagon driver very excited-

Vignesh Krishnakumar 1:20:10 Yeah

Speaker 1:20:10 ... to, uh, this, uh, CTO, who's thinking about industry problems, niche problems, uh, product maps, uh, high-quality talent? I don't know, I, I don't know what you think about on a daily basis, [chuckles] but how has that personal journey been?

Vignesh Krishnakumar 1:20:23 I would say when we- when we are, um, thinking about, uh, starting up, uh, in our wildest dreams we would not have imagined everything that we would have to go through and, and what all that responsibility meant, um, uh, right? There, there's no way I would have imagined what I would be doing today five years back or 10 years back-

Speaker 1:20:47 Yeah

Vignesh Krishnakumar 1:20:47 ... or even two years back, uh, for that matter. Um, I think it is evolving, um, uh, continuously, and the thing that, uh... There are a few things that help me go through this evolution. Um, one is the purpose, uh, that you're working towards. Um, every day I wake up and go to work, I know that if HyperVerge does a really good job at building the organization that it wants to be, then the upliftment that we intend to make happen for these 300, 400 million people is a more real possibility.

Speaker 1:21:28 Right.

Vignesh Krishnakumar 1:21:28 Right? So, um, that purpose is something that, uh, we find inspiring, and, and personally, uh, that gives me a reason to, you know, wake up and go to work, even when it's not easy.

Speaker 1:21:43 Are you saying that it's not really possible to figure out what are the skills I want all the time, but if the vision and the reason to go to work is clear?

Vignesh Krishnakumar 1:21:54 In my case, in my case, it was like that. Uh, maybe there is a way, uh, but we constantly keep learning about what is relevant in this space. So with, with entrepreneurs who are, uh, 18 to 24 months ahead of us, or 36 months ahead of us, we, we try to spend time with them, uh, learning about what is- what should we prepare ourselves for in this, uh, transitions, uh, that they're going through.

Speaker 1:22:19 Right. So as you reached out to Sridhar, you must have reached out to many other entrepreneurs.

Vignesh Krishnakumar 1:22:24 Yeah.

Speaker 1:22:24 Uh, and as you reached out to Seshan sir earlier, so... Okay, I get the thread.

Vignesh Krishnakumar 1:22:28 Correct. So that, that is something that, uh, we go through. Um, the, the second thing that has been very helpful for me personally has been meditation. Uh, so every day, come what may, whatever, [chuckles] whatever be the situation, um, otherwise in life, I take, you know, those two hours to, uh, meditate.

Speaker 1:22:50 Mm.

Vignesh Krishnakumar 1:22:50 And it helps me, uh, develop equanimity to whatever is happening, right?

Speaker 1:22:56 Mm.

Vignesh Krishnakumar 1:22:56 So how can I be, um, happy, irrespective of whatever is happening externally? Then we don't end up going to work, uh, or doing anything else, uh, for seeking joy from it, but we are already joyful, and the work that we are doing is a manifestation of our joy in the work.

Speaker 1:23:21 And of course, HyperVerge is one of those unique offices which has a meditation room in, uh, and al- almost always has had, even when you were a smaller office.

Vignesh Krishnakumar 1:23:29 Yeah. Now I hope it is not unique and more people do it.

Speaker 1:23:32 Yeah.

Vignesh Krishnakumar 1:23:33 That is our wish.

Speaker 1:23:34 That's amazing. So, uh,

Speaker 1:23:37 what I'm sensing is that all of you are very aware of the good fortune that you have had, and the effort that everybody else has put in into, uh, sort of making you who you are, and, uh, you've reached out to mentors, you've, you've spoken about, uh, the professors that taught you, and so on. And, um, you're very aware that to build a conscious business or, or there is something called a conscious business, and there is something that, you know, you're, you're building towards, um, uh, growing your revenue so that you can do something else with that revenue, so that you are in a position to influence, uh, the world in a better way, right? So, and what I also know of HyperVerge is that this is not just talk, and that you actually have a, a contribution wing. So do you... And, uh, and everyone you talk to in HyperVerge is very deeply connected with it. So do you want to talk about it and explain to-

Vignesh Krishnakumar 1:24:33 Yeah

Speaker 1:24:34 ... everyone?

Vignesh Krishnakumar 1:24:34 Yeah. So, uh, we did not want this to just be a vision or a dream, right? That we will become very large, have a lot of money, and then we will do something. So, um, uh, our, um, approach was always that as soon as we hit product market fit, and we are generating, uh, revenues and we are, um, out of survival mode, uh, we should immediately launch this. Um, so that is exactly what we did. We have built something called HyperVerge Academy, uh, where, um, we take people from underprivileged backgrounds and help take them through a six-month boot camp, upskill them on essential tech skills-... and help them get placed with at least a five LPA job in the tech industry at the end of this. We have a dedicated contribution team that, uh, runs this initiative, and every person in HyperVerge figures out, uh, some way to contribute to this, right? Um, so if you take people who are good at, at their own skills, so someone who's a software developer, so they contribute as a mentor in a full stack development, uh, course, right? So they take, um, maybe four or five students and, uh, mentor them on a daily basis, on a weekly basis, and, uh, be available for support throughout this boot camp, uh, to ensure that the student develops the skill set are necessary, has a mentor to interact with whenever they have doubts, and they are able to get this job. This happened in software engineering, this happened in, uh, AI teams, uh, it has happened in other places. We've launched a design boot camp recently. Um, QA is another, uh, course that we conducted. So there are, like, multiple of these streams that people can sign up, depending on their, uh, background areas that, uh, they are good at, they are passionate about, and we train them in that area and help them get a job. Then, there are people who help, uh, these learners with power skills. A lot of them might have the core skill, uh, but they may not be comfortable communicating, um, uh, with another person, right? They, uh, you know, they end up getting rejected in a group discussion or in a HR round or something like that. So how can we help them communicate? How can we help them be confident, um, in projecting themselves, in having a conversation, in reaching out, uh, when they have a doubt? So confidence is another area that, that we help them with. Uh, conducting mock interviews, uh, doing resume writing. How can they collaborate with other people, uh, well, um, in an environment? That becomes a very essential area because we can help them clear an interview and get a job, but what happens after they get a job, right? It is, it is very important, and they, they might have never worked, uh, in a collaborative environment before in, in their life, uh, at all in a, in a team project. So it becomes very, very important to, uh, help upskill them in those areas. So there are people who work on, um, power skills, um, with these learners. Uh, there are people who sponsor the, the program for, uh, these students, right? Apart from what HyperVerge is doing, they might do as an individual sponsor as well, additionally. Uh, people might need devices. Uh, uh, we, we try to ensure that it is not a liability on the family, right? So by default, uh, people who are coming from, uh, this kind of a background, uh, the trade-off that they end up making is if I, uh, take up a job, even some, uh, labor-intensive job or a, a data entry job or whatever, I am contributing something to the family, uh, versus if I take a course for upskilling, then I have to take some money from the family and provide account, right? So we ensure that, uh, people don't even have to do that trade-off, and that, uh, at least that much money that they would have earned is already given to them as a stipend. So the only thing that they need to do during the course is to focus on upskilling, uh, themselves. Uh, then there are people who build tools for this upskilling. So we've built a tool called Sensei, which leverages LLM. So when, when the mentor is not available, um, or when people are, uh, practicing in their own free time, et cetera, uh, we've built an LLM-based tool which, um, helps act like a mentor, uh, to the student, right? So they can pick up a topic like, um, "I want to learn Python," and, and it will break down exercises, uh, for these learners-

Speaker 1:29:34 Oh.

Vignesh Krishnakumar 1:29:34 - and it'll, um, ask questions on concepts. It'll give them programming tasks. It'll review their code, give it feedback. Uh, these are all things that have been built by people in HyperVerge, right? So, um, everyone, depending on their own, uh, uh, skills and volition, figures out some way to contribute to the system. Uh, today, where we stand, la- last year, for example, hundred such students, um, came through the system, got, uh, placed at the end of this, uh, with, you know, at least a five LPA job. Uh, and every year we are trying to compound this, uh, growth rate as well, right? The n- the number of people, uh, who come out of, uh, HyperVerge Academy, that should keep, uh, scaling as the, uh, economic engine grows as well.

Speaker 1:30:26 Nice. Um, we covered so much ground, uh, and, uh, we started with, uh, what HyperVerge does, and you spoke to us a little bit about how your AI is not a wrapper and is by built fundamentally ground up. Um, and almost through that entire journey, now we are talking about, uh, how HyperVerge imagines itself almost like a dual engine company with an economy engine and a contribution. That's amazing. Um,

Speaker 1:30:53 I want to understand what the next five, 10... I mean, if you w- wish to share it, what the next five, 10 years will look like?

Vignesh Krishnakumar 1:30:59 Yeah. So a lot of things are, um, uh, very dynamic from, from the life of a startup. [chuckles] So it's very hard to, uh, predict specifically what will happen. But few things, um, that are constant are things that I can talk about. Um, one, uh, from the company vision perspective, we want to help more learners, uh, go through HyperVerge Academy-... and as many people as possible from this three hundred, four hundred million people, uh, that we can take out of, uh, from below sustainable income levels to above sustainable income levels, uh, we would aspire to do that. So that is one trajectory that we'll keep, uh, continuing. Um, from our products perspective, to, to give an analogy, right? If you look at Amazon, um, see, a lot of things in the tech world continuously keep changing, but Amazon built their product vision around what is consumer behavior that doesn't change, right?

Speaker 1:32:03 Right.

Vignesh Krishnakumar 1:32:03 So as a-

Speaker 1:32:04 Always want it faster, always want it cheaper.

Vignesh Krishnakumar 1:32:06 Exactly, right. So you would always want things delivered to you faster and cheaper, and that is not going to change. It was relevant, uh, ten years before, relevant five years before, will be relevant five years from now, ten years from now, and the technology of making this happen can change, right? So imagine having an e-book store, uh, I mean, bookstore earlier that delivers books to you, versus having a Kindle reader, right? So, uh, and, and you just get an e-book, uh, delivered to you and start reading instantly. So, uh, the, the, the methodology of achieving the solution can change, but this specific consumer need itself remains. So in financial services, what we want to make happen is make financial transactions faster, easier, and more trusted.

Speaker 1:32:53 Nice.

Vignesh Krishnakumar 1:32:54 So, uh, all of these, uh, financial interactions that happen on a day-to-day basis, right? If you think about, uh, what should be the future of that ten years from now, people are going to want it to be risk-free. Uh, you don't want fraudulent things to happen. You want to be able to trust and do the financial interaction, uh, and you always want it to be faster and easier, right? So we will continue to build products that power this product vision-

Speaker 1:33:24 Yeah

Vignesh Krishnakumar 1:33:24 ... as a journey forward.

Speaker 1:33:26 Amazing. Thank you so much. I-- This-- It's amazing that you have built from here, like literally from this building-

Vignesh Krishnakumar 1:33:34 [chuckles]

Speaker 1:33:34 -where we are sitting, and we have... Uh, it's been, what, ten years, twelve years since?

Vignesh Krishnakumar 1:33:38 Ten years, sir.

Speaker 1:33:39 Yeah, amazing. I, I- I'm amazed at, uh, how much, uh, um... I mean, we, we say often that IIT Madras is the Best Place To Build, and companies like yours or Aether, they've sort of gone through that whole journey, right? Like CFI and then Incubation Cell, Research Park. Uh, you've had the benefit of great professors who have helped you. Um, most of you have done really well in your courses also, so there's a lot of academic... It, it's, it's beautiful. It's gorgeous.

Vignesh Krishnakumar 1:34:08 We're very fortunate that these things have happened, and extremely grateful for every person who has been part of this trajectory. Uh, and, and there are numerous people, and I'm sure I missed many, many, many names, uh, but grateful to the entire IITM ecosystem and every person who has been involved in this trajectory so far.

Speaker 1:34:27 Thank you, Vignesh.

Vignesh Krishnakumar 1:34:28 Thank you. Thanks, Amrit. [outro jingle]