Dhruv Sharma: Hey there, listeners. You're tuning in to TON 124. We're streaming live, and our first guest today is Saurabh Arora of Plum. Saurabh, welcome to the offline network.

Saurabh Arora (Plum): Thank you, Dhruv. It's so great to be here.

Utsav Somani: Welcome, Saurabh.

Dhruv Sharma: It's great to have you with us, Saurabh. You're here to talk about something new that you've launched. We'd love to hear it in your voice.

Saurabh Arora (Plum): Yeah. No. Thank you so much. Look. We've, we've come up with Plum AI Secure, which is like a suite of products for business risk insurance, and I'll give you a quick context on Plum before that. It's a it's a employee health insurance in a health care company, and we work with close to about 6,000 corporates. And two years back, we started a new business unit called Plum Business, catering to you know, we are, we are a IRDAI-certified broker, so we have the license. And, this time, we extended the business unit to cater to business risk insurance, and we have a distribution of all these corporates that we work with. So it was a good fit for us to use leverage this distribution and extend, brokerage and consulting, and new product building on business risk insurance as well. And this one, we've been watching for last few years, and we all know how transformative AI tech is. It's changing the way the software is built. Every company today is either building AI or embedding AI in its product or services or using AI to automate complete work. And the AI in the last couple of years has evolved from what used to be a scope pilot experience to now taking a lot more autonomous decision and increasing the autonomy and taking, a lot of decisions which could create new kind of what we call as business risks. And, we believe that the insurance also needs to evolve alongside the AI. So the Plum AI Secure Suite, starts with asking business on what is their failure mode. So we ask questions like, what is the AI doing in your business today? How can it fail? And what could go wrong? What could be harmed? And what are the financial consequences of those are? Once we have those answers, we bring together a suite of products, which is think about professional indemnity and tech errors and omissions where an AI enabled product or service is failing to deliver, you know, value and a customer alleges, sort of a financial loss to you. And in that case, a professional indemnity and errors, an omission comes into place. We bring together a cyber suite, which when the AI contributes to any sort of security, privacy, or a business interruption decision or incidents, this is when the cyber comes in, and we also bring together a D&O where a director and officers is facing a claim, on a governance or an oversight. So we bring, production around that as well. It's a pre underwritten product being negotiated. You choose between a 5 crore and a 10 crore offering, and straightaway go for the issuances.

Utsav Somani: I think this is fascinating. I have a question before I think, we let Dhruv continue the line of, thought. I mean so I read about this McKinsey report being McKinsey or one of the big four consulting firms, I think, used AI to deliver a report, and it was entirely, like, hallucinated output. Is that something that this kind of insurance covers?

Saurabh Arora (Plum): Yeah. I think the the the way to think about is that if that report was sold to their customers and those customers are challenging and suing them for sending them wrong information, and even worse if those customers have used those reports to make certain decisions in their businesses that have caused them harm. So the insurance suite is going to come into effect when there is a legal, repercussion or there is a financial loss that is coming because of that error.

Utsav Somani: Wow. Yeah.

Dhruv Sharma: When you conceived this product, Saurabh, I'm curious, who did you have in mind as as as a customer? Was it AI native start ups? Was it, you know, mid market or large enterprises who who are undergoing the AI transformation right now? Who did you have in mind?

Saurabh Arora (Plum): I think it started with us. We are, like, a full stack organization, which has it's a tech first company, but we are in an insurance business domain, which means, we have a large service teams. We have a large, accounts management teams. We have a fleet of sales teams as well. We have a full suite accounts based marketing and a marketing team as well. So it's an organization which is like a full stack organization. And we were seeing, areas where, we were using AI across the organization. And we are not an AI native organization. I can say that. You know, this is not a 10 people organization which started in the last couple of years. It's a 600 people organizations across multiple departments. So it definitely started looking us of how we were we as a tech company were evolving. But, you know, if you see where how the AI is evolving, every company is going to leverage AI. Every company is going to become AI company. Every company will embed AI, across their surfaces. It's such a transformative technology that AI is going to touch every organization. So whether you are an AI native company who is building AI or whether you are an AI, applied AI company who's leveraging AI, it's applicable for both of you.

Dhruv Sharma: Quick follow-up, Saurabh. Like, for an application layer AI company, what would failure modes look like? And then the other question to ask would be, can they realistically point fingers to someone else? Say, hey. We screwed up, but it was really the model that that started that whole thing in the first place.

Saurabh Arora (Plum): Yeah. Actually, there's a good precedent already set, on who takes the blame. But I'll I'll give you the some areas, you know, where this is applicable. So first area is, what if my AI makes a wrong decision? And a wrong decision could be a hallucinated advice, hallucinated answers. Even worse, if you are in a health care or a medical field, a wrong medical recommendation, and AI pricing errors. For example, you are coming to the websites or you're chatting with specialized chatbots, and you are getting wrong pricing of, or wrong refunding options in customer support. Think about incorrect financial analysis. You know, we have companies like Harway, Lagaro's, etcetera. Those are standing at really high risk. Those are really high risk industries by by nature as well. So, now, you know, our teams every team is using SDLC, like an AI led SDLC. So bugs introduced through AI coding agent, which is creating, some sort of disruption at customer's end. So these are all the areas where AI can go wrong, and the business consequences that the customer loses money and sues you. And and here, tech errors and omissions, a professional indemnity, could help. Second is a very important one is when AI creates a security incident. And, think about all sorts of prompt injections which AI could be exposed to. Today, we are giving large permission accesses of our databases and our warehouses to AIs. So think about sensitive data leakages. If you're a foundational model company and there are so many which is coming out of India, what if there are model weights, those are compromised and goes out in the open? So all of these constitutes what we call as security incidents and business consequences, incident responses, your legal cost, business disruptions, etcetera, and all of these are covered under the cyber insurance policies. And I can I can give you some more directions of, you know, how,

Utsav Somani: the interesting is Yeah? Are there any legal I mean, you mentioned some precedents as well. Has something been taken to courts in India?

Saurabh Arora (Plum): India, I think it's we are still developing. I won't say anything which is a hallucination risk that we are aware is already going on in the court. But as we as we speak, you know, they they this is just a matter of time. It'll definitely come out. Cyber insurances, you know, there are plenty of examples of how needs have happened in cyber insurances. A really interesting example, you know, close to home, not in India, but it's a Air Canada case. It's a a customer who goes to the Air Canada's website, is chatting with an AI bot, is interested in knowing about bereavement fares, and the bot responds back by saying that, everything is covered. The you could you could get refund of bereavements, when it comes to even on the normal tickets. And when the customer comes back actually to take the refund, the human customer support teams denies that saying that this is not as per the policy. So the customer takes them to the court. This is the Canada court, and this is a ruling of 2024. So and the court, the decision of the court is that because it's on the website. So they can't say because the Air Canada earlier was denying was saying that this is a third party AI that they were using, but the court, held the precedent that this is, the company's responsibility if it is on the company's website. So so that's a precedent which is you can't blame it on getting the efficient model

Utsav Somani: out of it, then they also should absorb the downside of it. So what did health insurance teach you about, building this? I mean, there must be many other things, but how do you underwrite something like this?

Saurabh Arora (Plum): Yeah. No. I think it's a it's for us, you know, some of the use cases are really important, especially when it comes to financial losses. And I'll give you an example. The third of our customer support queries, also we get is around our treatments covered or not. We process close to about 150,000 claims annually now, and a third of those claims have treatment doubts on admissibility and what kind of percentage, deductions will happen if there is any particular copays that they'll have to pay for particular treatments. And we've been trying to automate this operation for a while now. And today, 60% of our support operations are automated. And this is a piece where if our AI makes a mistake, where we say that this cataract surgery or this LASIK, surgery is covered in your policy, we have to pay it from our own pocket if we make a mistake. And, and I can tell you, we have made many, few in the beginning from our own pocket where customers have taken screenshot and shared this with us that, our AI bot hallucinated and said, this is admissible. So, it's definitely close to home, and especially when insurance claims, which is not like a 10,000 rupee ticket. These could be, like, tens of lakhs of tickets.

Dhruv Sharma: So speaking of this product, the the business risk insurance, like, AI risk insurance product, who should be paying attention to it? How early should they be getting this cover? Where can you know, what's the best way for them to get in touch with you? And once they get in touch with you, how quickly can you start covering them?

Saurabh Arora (Plum): Yeah. No. Thanks for asking that. I think it's, it's every technology leader or organization leader, who is leveraging AI and building solutions for their customers, or natively, you know, building AI themselves, they should start thinking what if my AI fails and where my AI can fail? And when the failure happens, what kind of, risks that they stand with? There are a lot of controls. You know, as technologists says, well, we can start building for a lot of controls, but for all residual risks in businesses when you're using AI, There's a plum secure suite. You could reach out to our website. We have a website where we are also explaining quite a lot for, you know, especially when we launched this campaign. You could go and engage with our website where we are educating you what does it mean to have, AI risks. And, we have a suite we have a team of consultants who are who are experts at this, who will understand your risks and give you, what we call as, like, custom fit products for your needs.

Dhruv Sharma: Have you guys pitched the Frontier Labs yet? Because off late, they have trouble keeping their agents inside sandboxes every day that that's paid.

Utsav Somani: Was yeah. My thoughts being where RBI was also holding meetings with banks and stuff. So was there regulatory announcements or internal movement, that sort of I I would be tailwinds for you?

Saurabh Arora (Plum): I I think this is definitely about to follow, which is, IRDAI has been tightening, the policies on how insurance intermediaries or insurance companies manages their own cyber risk. But what we are sitting ahead is a storm coming ahead of us, especially in the cyber incidents area and, you know, as the models get more more powerful. This this game of, Chord Police, which is, you know, the the attackers becoming a lot more powerful, and causing disruptions across our services is is just a time ticking waiting to happen. Our our regulators are generally for forthcoming and forward in putting out these regulations, and I and I hope that they'll bring out something soon. But, but the cyber, industry in general is super aware of what's coming ahead.

Utsav Somani: Alright, Saurabh. The risk is real. Thank you so much for building something for it. Thank you, and wishing you and team Plum all the best.

Saurabh Arora (Plum): Thank you, Take care. Thank you.

Utsav Somani: Alright, listeners. We're moving on to our next segment and guest, Vimal from SuperKalam. Vimal, welcome to the show.

Vimal Singh Rathore (SuperKalam): Thank you, sir. Thank you, Dhruv, for inviting me in. Glad to join.

Utsav Somani: Awesome. So let's start by introducing the business.

Vimal Singh Rathore (SuperKalam): Yeah. So at SuperKalam, like, as you know, I'm all in pro, in edtech since last decade or so, and, we are building AI personal tutor starting with UPSC enrollment exams. Recently, we have also launched for maths category in India. So, with the name is Homi of this of the second product. So now there are two AI tutor, Super Kalam, inspired by Dr. A. P. J. Abdul Kalam, and Homi for Maths inspired by Dr. Homi J. Bhabha. So that's what we are building.

Utsav Somani: And I think the question that you must face very often is how is this different from ChatGPT? Like, how does your app, like, help people or students prepare better?

Vimal Singh Rathore (SuperKalam): Yeah. I think, that was more pertinent, in in Homi category that is CBSE class 12 because they are it it is very difficult to become that, like, how to how to make a product that is good to have, to how it can become essential for the student's life. So there, we we actually are facing a lot of questions from parents as well, and we are answering those questions in in terms of product and features. But while we started with SuperKalam, it was surprisingly pretty much easier for us to navigate, this question. We rarely faced that why not GPT and Claude, although so many UPSC aspirants and so many students keep on using this. But when it comes to competitive examination, the trust becomes the first and foremost thing. That if I'm asking a academy question, whether the answer is 100% correct or not, there comes the importance of training, there comes the importance of reg that we have built, there comes the importance of community, competition. So when we give students a proper ecosystem where they can't just see themselves getting the query result, but other students trusting the ecosystem that gives a lot of coherent value, comprehensive value to the student. And I think in competitive examination, the FOMO is so much, the the outcome and the cycle is so small that it's just one cycle, one year in in most of the examination. That delivering outcome should be the only goal and that's what we have focused. That whether we are delivering the right value to the student, there should be least number of inaccuracy, least like, even now, there is a proper, system that we have built in terms of WhatsApp and Slack alert that whenever somebody is reporting that any specific MCQ we are right now, more than 1,000,000 MCQs are being solved per day, and more than 5,000 main answer, that is essay type answers are being evaluated on SuperKalam. And even if we get one, inaccuracy, we get Slack notification. After certain threshold, we get WhatsApp notification. And that's how we are able to build that trustworthy ecosystem that, you know, we are, becoming that platform that is go go to for the students.

Dhruv Sharma: So, Vimal, the UPSC means, if I'm not mistaken, are, like, nine days away. Yeah. Right? With less than ten days to go, how are students using, SuperKalam right now in the lead up to the exam? Are they writing subjective answers and getting getting them evaluated? Is that what they're doing?

Vimal Singh Rathore (SuperKalam): Yes. So what happens is, as you rightly pointed out, UPSC means it's just nine days away starting from August 21. And what happens, in that phase is that in five days, there are more than nine exams. The students write six hours per day. That is very, very competitive. Right? Now what happens since this is the second stage of examination where around 15,000 to 18,000 students appear, by nature, if the feedback is not given regularly, the students tend to practice more MCQs. They prepare more for prelims, rendering themselves less ready for the main examinations. That was the first problem, like you pointed out, Dhruv, that we started solving at SuperKalam, that instead of MCQ and academic queries, what is that one hero factor or one major problem that students are facing? So the pattern is like this, that prelims is just a filter a filtering state. The marks of prelims are not added to the merit. The marks of mains examination and interview are the only one which which which would get you which gets you the rank. But what happens is that if you are writing 20 questions, you know, even in a week, you you are waiting for at least two weeks, three weeks, or four weeks to get your answer sheet evaluated when exam is not there, when when you want to practice.

Dhruv Sharma: This was in the coaching centers. That's how It

Vimal Singh Rathore (SuperKalam): is in coaching center, whether online or offline. Right? So even if even if they have personal mentor there, they if they are famous, the bandwidth is very, very less to evaluate, or to give the personal evaluation. So if the feedback is given after four weeks, I'm learning my mistakes after four weeks, then my habit, my psychological inclination is not to write more answer, but to keep on revising, keep on reading current affairs, keep on preparing for prelims. So that's what we have changed. In just five minutes, for entire copy of 70 pages, SuperKalam is able to give answer, give feedback, not just in the form of, you know, English below the answer. It is it is giving on your own handwriting. So it actually makes a circle, point out the rate diagram, correct the spelling mistakes, tells you the concept that just like a regular human teacher would do. So that is the most loved feature. The most loved feature does not mean that it is most used because it is a difficult aspect of it. That is also very, you know, hidden aspect of UPSC preparation. Most used feature is still academic queries shared with the book, MCQs. But students who are getting selected and they have this feeling that they are in top 1% because 1% are being selected for the means, they tend to use this mains answer evaluation. So it is an attractive feature, but eventually, they go for the entire ecosystem.

Dhruv Sharma: And would would you say that AI gives you the opportunity to, in a sense, personalize the lesson plan for the student because the way coaching centers would do it, I would imagine, is just take a test and then segregate students into batches, right, which is Yeah. Efficient, but it's not personalized or hyper personalized.

Vimal Singh Rathore (SuperKalam): So, obviously, the answer, you know, you know, AI is giving us that opportunity for sure, but the deeper answer is that it is taking some time, because in education, especially in India, the functional product is more important rather than just personalization. Students just quickly want that because, see, students are not caring whether you are a AI product, offline coaching, or online coaching. They want result. They want quick satisfaction there whether whether they are progressing in their learning journey or not. So while, yes, technically, it is possible, but, practically, in the product journey, it is a very slow process to make student learn that how personalization in a tech product works. So what we are doing instead is that very simple personalization that okay. This is what you have read in the last week. On that basis, this is your quiz. On your sort sort analysis, on the basis of your progress, this is your quiz. So this is micro personalization, which is actually forming a lot of good habits among aspirants. Then we are bringing this concept. We it is already in pilot called Skyline and Highline where students are solving x number of questions in a subject, and they are getting the sense of achievement that, okay, if I do this, my my this percentage completion at this much accuracy, and I will achieve this much rank in the leaderboard. So these micro personalization, then we are going to bring that full personalization. Okay. This is a journey in the last one week. You press this button and just like you might have seen in wondering app or so, you will, you will get entire lesson plan, video lessons, etcetera. But perhaps the students are not very much inclined right away. They want they are maturing as as much as the product is providing them the functional aspect of it. But to answer, yeah, but to answer short, like, yes, personalization is pretty much possible. Technically, it is very much possible.

Utsav Somani: I think the one question that I had was one phrase that you use, chat with books. So you have acquired IP from other partners?

Vimal Singh Rathore (SuperKalam): So in UPSC, what happens, there are a lot of standard books, mostly NCERTs. So NCERT are free. Go like, there are NCERTs. There are state book books, like, from Maharashtra board, from Tamil Nadu board. So we have used that used those books, which cover almost 100% of the syllabus. Plus, there is a feature where you can upload a small PDF of any book, a small, you know, photo of any any any book or newspaper and ask the concept. Ask any doubt. So that is that is what we are calling a chat with a book.

Utsav Somani: But do these book publishers wanna partner with you and get their content out to students? Is that a Yes. Stream that you will explore?

Vimal Singh Rathore (SuperKalam): Yes. Yes. They are they are reaching out to us.

Dhruv Sharma: Fantastic. And, Dimit, tell us a little more about Homi as well. Like, who's using it and and how how young are those kids or students rather?

Vimal Singh Rathore (SuperKalam): Yeah. So, Homi is our very, like, pretty much dream project. We actually wanted to start our company, in in k twelve, but we know, with data and experience that k twelve is a suicide, that if you just start with that, you get these doubts whether to target parents, to target students, and how to bring that value. So we we wanted to start with something much more stable and then move towards Homi. So in Homi, right now, we have started with just one board that is CBSE, just one subject that is mathematics from class six to tenth. The problem that we are solving is pretty much what we have learned from SuperKalam that if like, we have seen companies in the past like Doubtnut, which have been pretty good at that time, and I think it was, it was before the time kind of product. So we have doubled down on the similar thesis, but with much more personalization. And for a small kid, what is the meaning of personalization? That if I'm solving a math question, can somebody tell me at what step I did this question wrong? Right? So that is the first thing. And it can mark in the copy itself just like we have learned on SuperKalam. So same thing we are doing it. Let's say if somebody is solving a question of mensuration or trigonometry, Homi comes and mark exactly okay or rectangle you have marked this angle a a b a a as wrong. This step is right as per CBSE or these these these many marks will get deducted. So this is what we have done. Plus, now we have launched new feature, the similar line. You can scan a photo and scan a question and ask model solution. We are now now launching games as well. So taking inspiration of whatever has worked so far, internationally as well for mathematics category and implement it in a way that find its resonance, with Indian market, Indian students. So that's what we are doing there.

Utsav Somani: There's one relevant, YouTube question that I wanna pick up from Alexandra. What are your thoughts about competitive exams? India is very big on them. Akash had said on the show that it's a flawed model. Do you agree?

Vimal Singh Rathore (SuperKalam): Fundamentally, fundamentally, even I keep on telling my students, in every interaction that do not prepare for any competitive examination, especially UPSC for more than two or three years. Yes. Fundamentally, it is flawed, but, you know, SuperKalam comes into picture when you have decided that, okay, you want to want to, you know, you know, swim this river, want to cross this channel, and that's where we come into picture. But very honestly, very practically, there are multiple. They're just a very long debate. But I personally think that, yeah, this, quantitative examination scenario is a little bit flawed. And with the recent, you know, phenomena in India, you we already know about, you know, how things are going on.

Utsav Somani: What is your take on the changes that will come from these, discussions that are happening out in the open now?

Vimal Singh Rathore (SuperKalam): I think not just these discussion, but this era. Like, when I was graduating from my school, you know, only only doctor, engineer, and CEO were the three professions. Right now, I'm seeing in my own circle, people are earning more than two lets, three lets, two, ten lets a month by just being a creator. And not just a normal creator, but proper knowledge creator, you know, doing something which is actually giving employment to, to their audiences. So being a creator with AI, giving so much learning out in the open, like, it is completely democratized. Like, today, if I'm I am from an engineering background, I want to learn everything about microbiology. It's pretty much one month task for me to learn something about it and build maybe a product or some industrial process around it. So I think what Elon Musk did twenty years ago, you know, popularizing semantic learning and, you know, going into hyperloop games and space, etcetera. Now pretty much every kid in India and the world can quickly learn stuff and and and and, you know, do their thing. I think it's it will be good. And I keep on trying to communicate this to my family, friends, and students as well that it is good that if we start depending on ourselves, even I give this as an example that let's say if, you know, next day, a supercalam or home is not there and my identity has been taken away from me, but my mind is the same. How would I restart my life? So I try to answer this question once again, that would I be able to, you know, achieve whatever small things I have done in my life once again or not? I think that is pretty much easier. You know, these days, we just need to, probably, probably do away go away with that past twenty, thirty years of learning that probably we have we has we are also giving to the next generation, you know, prepare hard for j, prepare hard for UPSC, etcetera. I think that is gradually that will gradually go down, and, people will do good in in in different scenario. I think knowledge, access to knowledge, such simple, easy, inexpensive access to knowledge is definitely changing things, and we are we are seeing this all around. So I'm pretty much hopeful and very much excited, like, me as a as an individual plus, as a person who is who is seeing these 20 years old doing such an incredible job, out there. And I think they will create a next ripple of, inspiration to the to the further next generation. So I think competitive examinations are necessary, but they will remain, like it's like multiple options will open. I'm not telling that competition in this examination will will decrease or enrollment will decrease. No. I don't think that is going to happen in the next two, three decades or so. But, definitely, people will be more hopeful about exploring things once they are not clearing examination or not choosing that, competitive examination routine.

Dhruv Sharma: Can you talk to us a little bit about the cost side of the equation as well? I'd I'd love to understand the the the comparison, right, between a student who comes all the way to Delhi. I don't know, stays in Mukherjee Nagar or one of those other places and then prepare for one, two, three years versus someone who can maybe stay back in their hometown and use SuperKalam to get prepped up and then even qualify.

Vimal Singh Rathore (SuperKalam): Yeah. Like, this I've been so I have been a teacher as well. I started my journey, from an academy, and, well, apart from leading their growth and strategy, I was a, like, continuous teacher. And till 2021, I was teaching in some way or the other. So and again, I have been a student of these examinations and further other examination as well. And I sincerely know with, like, with my own experience that self study is perhaps the most important part that forms, our our understanding of of any examination, especially when we talk about UPSC. UPSC is nothing but multiple easier subjects, which are too much in number, constrained with time limit, and many smart people are computing. So when the vendor is a time limit so it becomes the issue of time management rather than intelligence. So when you have more time, you are not going to coaching, perhaps this is good for you. This is blessing in disguise. And, you know, my wife has cleared, you know, UPSC, and many of my friends are in services without any coaching. There is, like, no coaching at all. I keep on meeting those students. While I'm not completely against anybody who feel, you know, in their own awareness that probably they have, they they feel that they are not so much aware about the subjects that are coming in UPSC at any point of time, they can go for any of them or or or online coaching. But I think self study forms an important part. With this with all this understanding, I started my first company, Coursavy. We decreased the price to around $1,517,000. But when when GPT launched in November 2022, and we were we were into YC, Suddenly, there were this there were this crazy night, where all all the three cofounders were sitting in Koramangala, Bangalore, and we realized that I think this is this is not like third wave of tech. We had seen NFT wave, we had seen the crypto wave, which a normal Indian was not understanding that why people are buying these, you know, photos for so many million dollars. But GPT, their responses, everybody was understanding. So we had realized that this is something which can probably make the unit economics much better because humans were sitting at the center before the AI era, whether it is content creation, logistics, live classes, people becoming hero, and then coaching was going on. It was pretty, pretty crazy stuff back in 02/2021 that all of us has seen in terms of education market or so. But AI suddenly gave us this opportunity not just to decrease the content cost. It is like a superintelligence is available, in in front of you knowing all the subjects. Now it's up to you. How intelligently can you can you morph the the UI layer which becomes super simple for any users to understand and, you know, extract it. So that was the feeling. And with that feeling, with that excitement, even today, we keep on pondering that, you know, what should be that next form factor, etcetera. And that that you remove

Utsav Somani: that brings me to my closing question as well because I think that's leading into that. That I mean, there's a standing critique of, AI tutoring apps where people need to find that motivation within themselves, and the completion rates have not been that great for self, led learning. Like, so how how do you think about that when you design, these workflows at SuperKalam?

Vimal Singh Rathore (SuperKalam): Yeah. Interesting. So, interestingly, retention and learning outcome was the problem that I had solved in my previous company, Coursavy. That was the ignition point to start that company, and and I scaled the same, you know, solution here as well. Today, not great, but our retention is 26%, day 30 retention, which if you see consumer app, which are only AI, is a is like four x or five x. I would want it to be around 40%. We are trying to go there. But I have seen consumer app at one or 3%, 4%. Right? People are getting selected, from our product. People are solving 1,000,000 questions per day. So retention problem is there from our perspective because we want to make it better. I think it's, and also it becomes the I think we are very lucky that, you know, we have a team where teachers are there, where operators are there, where builders are there. And me, as a single person, have seen all of these three in different different phases. So that gives a unique insight, unique nuances that, you know, that vice c phenomena that build something that don't scale. We really did that for the first two point five years. We are into scaling right now just from that last three or four months. We remained at some $20.30 lakhs of revenue for more than a year and kept on solving the unit economics. You won't imagine, I'm just disclosing a number. In in in in May, we spent 27 lakhs in marketing, and our revenue was 32 lakhs. In June, we spent 0 in marketing, and our revenue was 26 lakhs. And it was an experiment that we have been doing that, you know, you know, how to make how to make things work scalably. Because, see, that UPI autopay and that, you know, bluffing user into just getting their money deducted, we were never into it. But since since that is a phenomenon that is working out as well, we tried, and it it did not work because the DNA is is not there. Yeah. So that's probably answer your question that it is being as closer to the user. So two phenomena of two two two two things of YC that build thing, that don't scale. That were the first two, three parts of it. Now the second phenomena makes something people want. They just keep on calling user. I keep on interacting with our users with one way or the other, with their Telegram, YouTube, email or call every single day. So that really helps because it they know everything, what to build next, but prioritizing its importance, its depth that only comes from the users. Because generation is changing. For us, maybe twenty eight minutes video made sense, but for this generation, even ten minutes video does not make sense. So we need to understand that need. So that that that depth, we are pretty much enjoying. Everybody's enjoying, you know, that part. In terms of costing, SuperKalam is around one twentieth, one twenty fifth. And it is it is comprehensive. Even if you have any coaching, SuperKalam is helpful. You don't have any coaching, SuperKalam is helpful. And we have all the different kinds of aspirants. We have aspirants who have taken offline coaching, aspirants aspirants who have taken online coaching, aspirants who have never took coaching. And the percentage difference is not so huge that I should quote, okay, you know, this is the higher category. We have equal mix. Delhi, in fact, Delhi being the top city, you know, from where most users are coming, which is completely counterintuitive because all the coachings are in Delhi. Ideally, it should not be the highest city, but it is it is there. So I think very much excited, you know, more than just this category, building something that people are using on daily basis, building something that we had led in our nation. You know, we rarely see apart from Duolingo and Brilliant, we are not able to quote even third product in NetHack globally. Right? If you remove the and if you remove the language learning category, they're literally literally just brilliant out there. So it's a it's a huge challenge, and I think we are rightly placed, to solve this problem because we have been patient. We are not in a hurry. We want to solve this problem, and we really love

Utsav Somani: I think I'll have to cut the answer a little short because our next guest has been waiting for a while. Sure. Sure. No. But wishing you all the best, and thank you so much for educating us about, this industry that you're building in.

Vimal Singh Rathore (SuperKalam): Thank you. Thank you, Utsav. Thank you, Dhruv. All the best. Yeah.

Utsav Somani: Cheers. Alright, listeners. Third and final guest for today, Divanshu of Solinas. Divanshu, welcome to the show.

Divanshu Kumar (Solinas Integrity): Hi. Hi, Utsav. Hi, Dhruv. Thank you so much for having me.

Utsav Somani: Sorry to keep you waiting.

Divanshu Kumar (Solinas Integrity): No. Wait. There is at all. Yeah.

Utsav Somani: So let's, like, start with that.

Divanshu Kumar (Solinas Integrity): You want me to do is talk about what we're doing as a company?

Utsav Somani: If you can spend maybe two to three minutes introducing the company, the products, various lines of businesses that you're doing.

Divanshu Kumar (Solinas Integrity): So, we are solving the problem of underground pipelines, in the real estate, like, these cities or any commercial real estate for that matter. If you look at any, infrastructure, it has underground set of water pipeline, sewerage, and drainages, and so on, which we lay down, but we don't have right technologies to maintain it. Right? For example, if you have a problem of water leakage or contamination in a city that multiple cities face, finding where the problem source could be is very challenging. We dig the roads, do multiple trial and error rates, and so on. It takes multiple days for us to identify where the problem could be. We felt that technology could solve the problem. So what we have done at Solinas is we have built robotics and AI solutions that can actually help us understand how these, infrastructure works. So first thing that we do is we have robot for diagnostics where we understand what is going inside these pipelines, where the leakages could be, contamination could be, where the blockages could be, and so on. What is the condition of these pipelines when they're likely to fail? The second set of technology that we've built is for cleaning of these networks. So in sewerage and drainage network, we have a lot of problems of deserting, manual scavenging, and so on where people sometimes have to go and clean these systems. So we built a lot for that. And the third thing that we have done is we have built a digital layer of pipelines where we understand the entire network, bring it on our digital system.

Utsav Somani: I think some Internet issue. Right?

Divanshu Kumar (Solinas Integrity): It takes based on the data that we and we cut

Utsav Somani: The answer sorry. I think we lost you maybe for the last twenty seconds.

Divanshu Kumar (Solinas Integrity): Okay. So I was thinking that we have built a digital layer through which we are able to get everything on a platform, and and we're able to run analytics on top of that. So that's the way that we have built the entire, layer of, end to end system for managing underground systems. And, we have done this, across more than 30 cities so far. And we are today one of the actually, the largest player which can do integrated technology management for all the underground systems.

Dhruv Sharma: And, Devanshu, when you do this, are you typically talking to and selling to civic authorities, municipalities, or or builders, RWAs? How does it work?

Divanshu Kumar (Solinas Integrity): Sure. Sure. So, yeah, as you can see from a volume perspective, the largest set of customer that we'll have will be the municipalities, the civic authorities who will be there. But what we're also seeing over the years is a large chunk of these, sort of cities, their water and sewer management is getting privatized. So on one side, we have a lot of private operators who become our customers. So for example, there are company like SUEZ. It's a French company. There's a company like Veolia. L and T is emerging as one of the major place in this segment. And a lot of other industries are also setting up water businesses who are taking interest in running these operations. So today, about 55% of our business comes directly from the government where we work with the utilities directly. About 25% of them, the remaining would come from, part of partnering with these private utility operators who in turn work for the com the governments, but we have a separate contract with them. And some part, we also work with industries. Great line of business as well where we work with them for the limited underground infrastructure that they have.

Utsav Somani: And I mean, so I mean, the aspect that you mentioned basically, the AI and the physical aspect, like, how did you go about collecting this data? Because you started in 2018, and even the origin story that you, I think I've highlighted somewhere in the Harvard Business case study also is fairly interesting on how you entered this industry. But how has the product line evolved since, I mean, AI is upon us now?

Divanshu Kumar (Solinas Integrity): Sure. Sure. So while we registered us to say 2018, the actual execution only started in 2021. Because in the first, like, 1819, we were still in college. We're trying to do something. We worked on certain r and d then COVID hit. And, then, of course, we went almost went down to zero. So we all also discussed whether we wanted to continue running this company or not because there was nothing else. We had not raised any funding, and there was no customer available for us. And as you can imagine, like, hardware, you cannot do remote working. Right? So we had to rebuild the entire team at the end of twenty twenty, and that's when the actual journey of Solinas, as I call it, Solinas two point o started. And, initially, the first part that we had was just pipeline diagnostics. Right? That is what we started with. The product of cleaning, manholes and serial networks, that was still under r and d. That that's something that I've worked on as a part of my final year project in college. So the first product that we launched in early twenty twenty one, that was the pipeline inspection one. We were just collecting data. Wherever somebody would have problems in their pipeline, they would call us to give us small small projects there. That is how we started. Right? That time, even though water diagnostics piece was not something that we had worked upon, only when we did a lot of work in 2021, people said that apart from c one inspection, water inspection is very key. Because in India, one, thing that I will talk about as a background. In US or Europe, you can directly drink from the tap because water is running in these pipelines twenty four seven. It is pressurized. But in most of the developing countries and countries like India, we only get water for two hours. So the water that you drink from the tap is today undrinkable. And because of that intermittent supply that happens in the pipelines, there are much more problems that happen, of each leak contamination and so on. And even when the pipeline is leaking, there is no pressure puff that comes out. Right? So it just seeps through the ground. And that's when the biggest challenge of diagnostics come in. That you know that the pipeline of water is contaminated. But where exactly it is getting contaminated, nobody has any clue. We saw a major disruption of that in an unfortunate incident in Indore, If you have studied that, where almost, like, sort of fifty to hundred people lost their lives because of that. Because the whole area was contaminated, and we couldn't figure out where the contamination was coming from. So what the your ability operators told us is that if you can build a smaller robot that can go into these smaller diameter pipelines, it would be of a good help to us. So that's when our team started doing the r and d, and we built our robot called Endo90. So Endo90 is technically Asia's first robot that go into 90 millimeter pipelines, which is generally the smallest distribution network that is there. And we send the robot inside, and it, kind of it's really like an endoscopy. Right? So it will go inside and it will look at, what is there, whether it's a leakages, contamination, there is something else, or so on.

Dhruv Sharma: Can you talk to us a little bit about, Devanshu, like, the the conditions, like, the real world conditions that make it really hard for endo ninety to operate because, obviously, it's dark. I'm guessing it's a very corrosive environment. There must be obstacles along the way. So talk to people about all of those things as well.

Divanshu Kumar (Solinas Integrity): Sure. Actually, if you see the real condition of the water pipelines, you will, have trouble drinking water. So that is there. You can imagine everything that can go in. Right? We have seen as much as live rats inside the water pipelines, right, that have become a source of contamination in that, and we have, like, solved for that. But, typically, what makes it very much difficult is because it is intermittent supply. And for the two hour that water is getting supplied, remaining twenty two hours, the water is just stagnant in the network. And any stagnant water leads to a lot of sediments. So the pipeline start getting sedimented and getting deteriorated much, much faster than what it would do normally. So one problem is that that, one fifty millimeter diameter pipeline can essentially become one twenty, 110 over a period of time. Two, a lot of people actually put in, illegal corrections into the pipeline to get more water, and they damage the pipeline like crazy. Right? Sometimes you will see every one meter, there is a ferrule that is being inserted that has damaged the pipeline. So that also the robots have to navigate. Sometimes there will be times that between the two pipes, there will not be anything. So, like, people have forgot to connect the two pipelines itself. And a lot of these are because of contracting sort of, challenges that are there. So all of these things make it much more difficult for us to operate there. And that's why now, even though we built the Endo90, now we are further doing the RND to build a micro robot, which can go into 70 millimeter pipelines. Because we realized that while the actual diameter of the pipeline could be 90, after the sedimentation and scaling, it ends up becoming, like, a 75, 70 millimeter. So with the smaller robot, we can go into these pipelines and ensure that disruptions will not happen.

Utsav Somani: There was an article, I think, which said that you've helped the Bangalore authorities avoid over I mean, close to 400 cases of digging.

Divanshu Kumar (Solinas Integrity): Yeah. Yeah. Yeah.

Utsav Somani: So, I mean, they must have saved a lot of money and in including the citizens who must have avoided so many headaches. So

Divanshu Kumar (Solinas Integrity): Give a comparison.

Utsav Somani: Yeah. Sorry.

Divanshu Kumar (Solinas Integrity): Please please go and finish your question.

Utsav Somani: Do you charge your customers on how do you charge your customers? Is it, like, just a subscription per, project task or, like, is it something else?

Divanshu Kumar (Solinas Integrity): So we have two models of business that we have. One is where we sell the product to them. And then if needed, we take the operations of that robot as well. Right? Where the this thing would be a monthly contact and so on. Second is when we provide they don't have to purchase the robot. We provide end to end service on a monthly basis because these are requirements that come on a regular basis. City like Hyderabad, for example, will have 60 to 70 complaints a day of water. Bangalore would have about 40 to 50 or maybe more depending on sort of some time of the day or the summer season and so on. So how quickly you are able to diagnose the problem becomes a very important KPI for you to solve the problem. Because once you know that at this location, 30 meter from this place, the problem is there, then it is just a half a day job for you to dig that road, cut the pipeline, replace that, and move on. Right? But the critical challenge is diagnose diagnosing that thing. And for that, people do hit and rise. So I'll tell you the parameters that we work on. Before our technology, on an average for one diagnostics, people take about five to eight digs to identify where the problem could be. We move that to a single dig from where the robot enters. That's all. So that's a five to eight x multi plan that we create. What would otherwise take about three days for them to diagnose a thing, we bring that to less than two hours. Because all you need to do is just send the robot inside, and in, like, sort of half an hour or one hour, it'll tell you where the problem could be coming from. So the multiplier that we're able to create in water diagnostics is immense. And today, we are live across 18 to 20 cities across the country where this is being used, and all the places they have used are robots. And I think the scale of impact and leverage that we're able to create is phenomenal.

Dhruv Sharma: And I'm just wondering I mean, you clearly identify yourself as a water diagnostics company and and, you know, talk about what water and sanitation, those two areas. But if you I don't know. If you've ever thought of yourself as a pipeline inspection company, then, you know, there's always the oil and, natural gas industry. I'm I'm sure you get this question as well. Yeah. Yeah. Tell us about why you chose this specific why why you chose to stay with this specific problem set?

Divanshu Kumar (Solinas Integrity): Good. I think even early on, we had this question about whether we want to be a general diagnostics company or we want to be a bottle or wash company as such. I think the answer to that comes from two lens. Right? One is in the origin of why we started in the first place. Right? This was a problem that was deeply personal to me as well. Largely because of the way that I have also grown up in a sort of tier three town, tier two town. Like, I come from KIA in Bihar, right, where bottled water was a luxury. Like, we would only get bottled water where somebody as a key guest would come in. Otherwise, by default was that you open the tap and you drink from that. Right? So seeing that transition where today, I can't have water without, like, sort of a bottled water we present is a great addition. I know that it's also problem of access for a lot of people. Right? If you think about all the people who, go to hospitals or who have this public health issues because of breaking water, all of them are people below the poverty line. Right? So it is definitely about access to quality water. That's a basic need. So that was always there in the head. Other thing why I got into the sector was because of my final year project where where we were building the robot for cleaning the manholes and sewers. And that also has a huge impact on how I would see these problems, like, sort of translating across hierarchies as such. Right? So the first part of the answer is purely about the reason we started the company, which was to solve the problem of water and sanitation. Second, I think also, I see there's a good white space that is there where we as an early entrance could create a market. Right? We are able to shape a lot of contracts policies in the system because we are early movers. Right? And that definitely also gives you a business advantage. I personally feel that of course, maybe three years, five years down the line, we'll expand to other segments as well, maybe oil and gas, defense, and so on. But being in water also gives us a unique advantage because we have persevered for the last five years. Now people know us as a company that they can trust. In any city now when a problem happens, largely, I would directly get a call from the municipal commissioner or the secretary there and say that, can we discuss with you how do we solve this problem? That's an advantage that you have. The brand that we have built as somebody who cares about the problem and is not just trying to make a quick buck, that's very, very important for a long term business outcome as well. And then I feel that it puts us in a good position where we are able to solve a very large problem that affects billions of people across the country, as well as make a good, business out of that.

Utsav Somani: Congrats on the 5,500,000 that you've just announced, I think, in August, last week, actually.

Divanshu Kumar (Solinas Integrity): Last week.

Utsav Somani: Yes. And you've also gone live in Dubai. And I think the other thing I mean, so this money will take you, of course, to, I mean, much greater heights. But what is your next set of challenges that you wanna tackle? Is it mostly hardware? Is it mostly software as you think about new industries like Dhruv mentioned oil and gas and other, related, sectors?

Divanshu Kumar (Solinas Integrity): I think the larger challenge would still be the scalability that we want to build. Because, the potential is that we can become, let's say, a thousand, 2,000 crores company in terms of revenue. But today, we are still, like, sort of describe the surface. Right? So I think the larger time would still be about, how do we scale that business to a certain height that we want. And I I really believe that within India, we have a great potential. Right? Second would be, I think, the various indigenous are They're able to work on. Because the problem like, you can't have a robot that works in Europe. Copy that in India and sort of try to write. It will never work. Right? So how are we consistently able to bring in new technology that solves a larger chunk of the problem? That would matter a lot. Right? So for example, one of the RNB that you're working on in hardware is building a robot for cleaning the stormwater drains so that flooding does not happen. Now on one side, it's a social problem. I've been able to see it. But I'll talk about the business opportunity, which is crazy. Just the top six to seven cities alone, Bombay, Delhi, Hyderabad, Chennai, Surat, Ahmedabad, and all of that, spend about 15 to 1,800 crores a year trying to clean these, sort of systems. And that also we know that a lot of time they fail significantly because the flooding problem does not really go away. Right? As we are building more and more cities on the outside, let's say, we're gonna have him by Gurugam, Fahedabad, and so on, this promise will continue. So it's a massive business opportunity that is sitting ahead of

Utsav Somani: sensors that can be installed. Instead of robots, maybe, like, sensors at a distance of, like, I don't know, a 100 meters, 200 meters, which are much cheaper and can be used for continuous monitoring to solve this

Divanshu Kumar (Solinas Integrity): problem. You can definitely do that. The problem is that the environment underground is, not is very rugged. Right? The sensors don't sustain. Like, so for example, one of the people one of the companies tried to to build manhole sensors. Under the manhole side, we tell them when the pipelines are overflowing. But if the pipelines are overflowing all the time, then that solution does not work. Right? So it's a lot of system problem that you should have enough equipments and systems to clean them on a regular basis in a faster manner so these overflows do not happen. So it's a larger chunk of the problem that is there. We believe that if we continue to innovate indigenously for these systems, we will continue to sort of create new markets and build our place there. And simultaneously, if we feel that there's a opportunity across oil and gas by sort of, kind of, modifying our technology there, we can do that. So to another example, like, when we're building these, drain cleaning technologies, we can customize some of them to actually do tank cleaning also for oil grease. Right? And, of course, we need to take ATX certification, safety, fire, and all of that. But if we're able to do that, that could give us a commercial opportunity there. So the focus always remains solving for, urban, sort of utilities. But simultaneously, if we can get into the adjacent sectors, we would love to do that as well.

Dhruv Sharma: Fantastic. Deviance, I'm I'm curious if you've, either heard of or read this book called Tube. It's, it's actually about, you know, undersea Internet cables. But because, because you work in the world of pipes, I don't know, might just be an interesting read.

Divanshu Kumar (Solinas Integrity): I have actually not, but just last week, somebody mentioned that to me. So I'm now very curious to see what is inside the I'll definitely order it today. Yeah.

Utsav Somani: Awesome. Devanshu, thank you so much for coming on my show. Congrats, and all the best again.

Divanshu Kumar (Solinas Integrity): Thank you. Thank you so much, sir. Thank you so much, Dhruv. It was a pleasure. Bye.

Utsav Somani: Alright. We've still got a few minutes before 05:00, so let's cover some news. There was a 6,500 word AI essay that Zuckerberg put out, and he's posting all of this on x. So could you get a chance to read it?

Dhruv Sharma: If so, please I I have it. 6,500 words is a lot of stuff, but I I I read through pieces of it, and then I have an AI summary, as well. I think what what what, let's just call him Zuck, what he's doing is really making a case for personal superintelligence. I think the alignment school of thought so far has been, like, you know, the Frontier Labs are gonna get ahead of everyone, and then power's gonna, in a sense, centralize in their hands. And and Zach is making a case that's counter to that. The other big point that he's making is that right now, everyone's talking about AI that, you know, using as a tool to automate. But, really, the way it becomes extremely useful to us in our everyday lives is when it's not just automating, but also inventing. So, again, invention but not automation. That's the other big, point he raises in the essay.

Utsav Somani: So I think I mean, he's put out manifestos before, but do you think he follows through on this?

Dhruv Sharma: Yeah. I mean, I think manifestos are also a strategy document in a sense you're also posturing. Many have said that this is also self serving argument because of his several multibillion user products and all of the surface area that that that you have personalized super there's nothing, more commercially savvy in the world than personalized super intelligence paired with advertising. So for those are all, like, those are I

Utsav Somani: mean, he spent a lot of money getting into the AI space, and I think he's still trying to play catch up. And there was some argument on Twitter which was going around. The counterargument to this is that I mean, he wants open rate models to be openly accessible to everyone. Can we all of these to be given US access as well? Yeah. And because he's been left behind in the closed AI race, I think there's a case that he's making that can we make it open? Can we distribute it widely? Can we all have personal intelligence? And there's a company that General Catalyst also funded called River, I think, which is trying to do something along these lines as well. Mhmm. That's one to check out. And there's Nvidia and Wall Street trying to collateralize GPUs now. Yeah.

Dhruv Sharma: Well. Someone's gonna pay for all of this. So I I I think, again, the underlying idea there is that in the end, investing is about, you know, sharing in on economic prosperity. And if AI is the future of economic prosperity, then we must find a way for, you know, everyday investors to, you know, to access it. So, NVIDIA has partnered with all the big, you know, financial institutions, many of whom are trying to, you know, bundle new products for retail, for retirees, and so on and so forth. So, yeah, that's the big $500,000,000,000 question.

Utsav Somani: 500,000,000,000. They signed memorandum, memorandums of understanding with six of the largest capital allocators in all, Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to compute financing platforms, aiming to mobilize more than 500,000,000,000.

Dhruv Sharma: It's gonna be like like REITs, but for data centers. I think that's the other way to understand this.

Utsav Somani: That's the other way. Yeah. So in basically, I mean, Jensen said that in AI computers revenue, the pitch is that a GPU is not a depreciating hardware, but productive infrastructure. So like a financial toll road of sorts, basically. And to answer the circular financing critics, Nvidia says it'll back stop up to 25% of residual value once selected projects.

Dhruv Sharma: Because they have an investment grade balance sheet, which is almost as strong as any of the other institutions that you mentioned.

Utsav Somani: And Axel, five fifty million India fund oversubscribed in weeks. And this is the ninth India fund, which was announced just today morning, and it's part of a coordinated 3,500,000,000 global raise. They've got a 1,350,000,000 growth fund, that can carry a company from seat to IPO. So more capital for Indian startups, and there's personal changes at OpenAI yet again. Brad Lightcap, who was the CEO, and before that, he was the CFO, eighth year and is the latest, person to exit from OpenAI, which just announced a 7,000,000,000, stock purchase program from their employees. So good liquidity all around.

Dhruv Sharma: Yeah. One hears of all of these departures. It's so I wonder, like, we don't hear too many announcements of people coming in and taking their place.

Utsav Somani: Maybe they come in, but I think, I mean, I read the article on this and there was this funny thing which said that, one person I mean, they, of course, didn't name it, but he was part of the interview, in that article. He said they cashed out over 10,000,000 and is now traveling the world and not working.

Dhruv Sharma: So it's like the like, cloud exits but silent entries almost.

Utsav Somani: Could be, I think, because there has been so much change. But brand light cap, I think, wants to start something new. That's the story that they're putting out. So I think, one more exciting venture to look out to.

Dhruv Sharma: One other piece of news is, I believe, the Tata Sons chair and Chandrasekaran has announced that he will only be with the group until, I think, February 2027. And so although that's not technically startup news, but, you know, Tata Sons affects our lives in so many ways. So that's, you know, pretty important news as well.

Utsav Somani: Alright, listeners. That's it from us. We'll see you on Friday at 4PM. Thank you so much for tuning in. Bye bye.