Building India’s Vision AI Ecosystem and the Journey of JARVIS

Speaker A: So good evening friends and welcome back to this episode of uh, AI Rising. Uh, I have of course my co host Jayant Kola, uh, who is not as ebullient as, or as effervescent as he typically is. I don't know what the reason is. We'll probably get to that in a bit. And then of course I also have uh, uh, Atul Rai with us. Uh, and I will you know, introduce Atul Rai in, in a minute. But, but uh, if you still don't know who he is, uh, have you uh, folks, you know, you must have heard about the Ironman and if there's, if you have heard about Iron man, you have surely heard about Jarvis. Now that Jarvis is J A R V I S or just a really very intelligent system. And that was Tony Stark's AI assistant. Originally a natural language system after the Stark family, uh, butler, uh, it evolved uh, to manage Stark Industries, secure Stark's properties and assist the Avengers in uh, combat. And then of course later Stark integrated Jarvis into an all uh, you know, Iron man suits. Ah, of course it was destroyed during the Ultron uh, offensive if those of you who watch these movies, uh, um, but sort um of you know, Stark later uh, merged the remnants with a synthetic body and an Infinity Stone to create, you know, Vision Jarvis, which was succeeded by Friday. FRI D A Y Now why am I telling you about Ironman and why am I telling you about Javis?

Speaker B: It's simply because, because you're a fanboy. Because you're a fanboy.

Speaker A: One is a fanboy. But also because you know, Atul Rai's company STIO Technologies has developed an advanced AI video analytics platform called Jarvis and recently he uh, sort of also implemented it or deployed it at the Vijay Bhoomi University. Now uh, Jarvis of course enables real time intrusion alerts, gender based access control, operational intelligence. The whole idea is to try to build a smart campus or the future ready campus, whatever you want to call it. Um, I think what uh, Jarvis, and which Atul will be telling us in a bit about it, it's his ability to provide real time surveillance alerts, actionable analytics, uh, and that's, that's pretty interesting. So you're more proactive rather than reactive. Now Atul co founded uh, you know he's the co founder and the CEO of uh, Stacku Technologies. He co founded this company in August 2015. So that's a uh, decade back along with Anurag Sahini who is now the CPO and co founder and Pankaj Sharma, uh, Who is the CTO and co founder Atul previously worked as a research consultant. Uh, Jain, this should make you happy. At the University of Headfordshire. And he also completed his master's degree in the. He's an Ms. Uh, of Science, uh, in Artificial intelligence from the University of Manchester. So uh, which year was that? Uh, Atul?

Speaker C: It was 2011 and 12. So I passed out in 12 and we joined the graphene lab for a while, uh, where I completed a master dissertation. So graphene is one which. For which the Manchester got the Nobel Prize.

Speaker A: Yeah, correct, Correct. Pretty pretty interesting. Pretty interesting. So 2011 is just pretty quite recent where as far as AI is concerned because given that it's a 60 year old technology, even machine learning is more than 40 years old. 2011 is pretty recent. So Atul is very, very well queued up as far as the AI. Just very close at just four years before the transformer technology. So pretty interesting times.

Speaker C: Uh, 21 on transformer itself. So when I publish papers. So when if you talk about my AI research thing. So I did it till 21. I mean my last research paper got published in 21.

Speaker A: Oh, so very much in the transformer stages. Absolutely transform. What 2017, I think 17. Yeah, around 2017.

Speaker B: All you need is attention.

Speaker A: Absolutely. Now folks, of course, um, when you speak about Stark View, it has got, you know, a range of sectors, petrochemicals, retailers, even state governments, all other, all of them are using Jarvis. And the dream of Atul, which he has stated in the media, uh, also is that every store, big or small should have its own Jarvis. I like the punch, it's a nice punchline. So I'm um, just quoting it verbasim and we'd be having quite a detailed uh, discussion with Atul. But before that, as is the practice with this uh, AI Rising podcast, Jayant and I usually have some discussions, uh, around the developments of the week. So folks, what happened during this week? Of course one is that Google introduced its AI mode in India too. It had these soft launches. Yeah, it had the soft launches with the labs, etc. You had to use the lab sign up. Now it's without the lab signups. Uh, AI M mode was already introduced in May and now in uh, June. So you have that meta is powerful, you know, powering its super intelligence lab and poaching people right, left and center from all the other companies. So OpenAI, uh, uh, Apple, uh, Google, everybody is going to have some trouble with that. All the senior guys.

Speaker B: Just this morning, OpenAI poached four guys from uh, Meta, um, Grok and um, I think Apple.

Speaker A: Yeah it did 11 guys earlier and Daniel Gross from OpenAI and uh, uh, from NFDG and all those kind of companies. So uh, it's crazy stuff. But the interesting part is out of those 11 hires that it did earlier all were migrants. So it's pretty interesting in what happened with Donald Trump and you know, getting migrants back to Meta, pretty interesting. Nobody lost that uh, angle, you know everybody, everybody latched on to that uh, very interesting thing. So I must say Zuckerberg has guts. I mean despite what you want to say about it. And of course um, the third development, which is again a very important development is I mean development in the field of AI is Grok 4. So Grok 4 is trying to, you know, I mean there's an announcement that's going to happen uh, uh, tonight uh, because uh, by the time this podcast is out, because the uh, announcement will come later so we know the details, probably will be discussing it in the next episode. But uh, that that should also be pretty uh, interesting. And uh, of course um, you know the uh, there's a nice, very interesting report that I have been writing about in my newsletter and that's you know, about the free and open source software. So uh, there's a report that has come out, the rise of force in India. Uh, it's done by the National Law School of India University uh where they mentioned that you know they have underscore this point which has been earlier also highlighted that FOSS components of real uh, free and open source uh software components now comprise 70 to 90% of all modern software systems. In fact actually I don't think most people know this but uh, almost, you know more than 85% of India's uh, Internet itself runs on force. Yeah. And then you have you know, companies like uh, I mean the digital payments uh company NPCI that runs on force. The uh, Kerala government of course started it out way back in 2000, uh, uh one onwards and other state governments have also picked it up. And you um, have Zerodha of course which has got its whole, you know, that order management system itself uh, on uh, free and open uh, source software. So pretty interesting uh, stuff. And I you know just uh, for the benefit of the. We have pointed this out earlier also but According to the AI UM Index 2025 report which is by Stanford in last year India was the second largest contributor to AI related GitHub projects. It accounted for nearly 20% of global contributions just behind the US at 23.4 and slightly ahead of Europe at 19.5%. And of course this trend, uh, signals India's growing influence in the open source AI community. So pretty interesting, uh, stuff. Uh, Atul, Jayant, any quick thoughts on any either of these, uh, developments?

Speaker C: So one thing, uh. Okay, Ajant.

Speaker B: No, go ahead, Atul, go ahead.

Speaker C: One thing which I have identified in this whole AI this thing is like there are two kind of a players in this market. One is who's working on the foundational AI architecture models and all. Second is more on the application layer of the AI. And of course both are important. But somehow in India we are quite behind that, uh, the trend which is the foundational model and all. But given that we are a country of engineers and software developers, the application layer is something where India can rule the world. I mean of course we are having a lot of engineers front end, back end, uh, DevOps and everything. So that is something that India sort of now utilize. It is strength. And that's why you will go with the git.

Speaker A: Right?

Speaker C: Lot of git projects are actually an application on the top of AI models. Right. We are not developing llama or Gro or uh, any m. Similar models.

Speaker B: Right.

Speaker C: So this is something good because ultimately we do not have that infrastructure or in very short thing, brief term, we do not have funding.

Speaker B: Patient fund.

Speaker C: We don't have a capital.

Speaker B: Patient capital. Patient capital.

Speaker C: And uh, investors are here. Like of course they can get um, uh, a return. The last time I was discussing with you while selling the ice cream and that got quoted by Pius Goyal also. Right. So of course if you sell ice cream, you get a return faster than making a model and generating a revenue out of it. And uh, that is what like India is for the investors. So making money.

Speaker A: Very interesting that you say that, but just in this specific context. Ashwini Vaishnav has been speaking about the Indic LLMs which you know are being touted as Foundation LLMs. I have written a whole piece on that explaining.

Speaker C: Right.

Speaker A: My own, you know, views on that. Even Jayant contributed to that. But I would love your views because you also, you know, uh, go with

Speaker C: all these indic LLMs and all the foundational model. Most of them have llama. Right. So uh, uh, I mean even in those models also you have to borrow it from Meta. So that is what the open source is doing. So I would not say like I'm not, I uh, mean I'm not putting that thing that most of people are not developing the foundation model. But most of the companies who so far are saying, of course if you go to their research papers or their git and everything, the foundation is lama. Which is where I still feel that India still need that research ecosystem. Just not a system. Uh, the ecosystem where you have academics, you have uh, startups, you have enterprise government, everyone is coordinating for that or vouching for that.

Speaker A: So do you think that the government allocated something like 1 lakh crore just recently that announcement? That will probably sort of help in this regard to some extent at least Government always.

Speaker C: There are two, ah, very important uh, contribution for any government in any of the world part when you go to the France, because I was part of India mission for the France collaboration.

Speaker B: Right.

Speaker C: So one thing which we've seen that there are two things. One is of course the capital and second is a very important part is sentiment. When government pushed that we want to do AI, then everyone around, whether the government officers, where the policymakers or where the enterprise, everyone then start following it up because everyone want to work in collaboration with the government. Right? So that is good because that's because startup, when you like remember 2015, startup India and all right, that point of time, startup was there, but at such, no um, government sentiment or support was there. But suddenly government started a program Startup India. I don't know how much it contributed directly to the startup world, but it created a sentiment where international investors, everyone started coming to India. So this is I believe will also go and help India in AI as well because Modi went to France and um, shown his interest with collaborating with uh, French research centers. And in that way even top 10, uh, or 10 AI starters being selected from India and sent to Paris. Three months there, right? We were there for three months. Right. So that is something I see. That's a sentiment. Once that sentiment get created, then money, capital, time, everyone of course is good. Of course invest into that. So that's something which is good.

Speaker A: Uh, Jayanth, your quick thoughts.

Speaker B: Yeah. Uh, on a few things that you already mentioned. First of all, let's, let's get back to uh, the 11 key hires that Meta, uh, hired in the senior AI team, the core AI team. I was just counting. Eight of them are from China and Indian descent. Chinese and Indian descent. One of them is.

Speaker A: One is from India.

Speaker B: Yeah, one is from India. Eight, uh, sorry, eight from China. Uh, sorry, seven from China.

Speaker A: Okay.

Speaker B: Chinese descent, one Eastern European and only two natives. And uh, as much as there's a geopolitical and an AI tech war between us and China, the key guys in US, uh, AI ecosystem come on for heaven's sake.

Speaker A: Nvidia and AMD are both.

Speaker B: Exactly. That's the point. Uh, getting back to your GitHub, uh, this one, that's another thing that I wanted to highlight. I think we've discussed enough about uh, you know, India's ranking, uh, and Indian engineers contribution into GitHub, uh, and the motivations behind that. I mean you know we know that India ranks number two consistently in all kinds of uh, AI competency, AI, uh, you know, upgradation and uh, talent pool. But in all these reports there are two things that I very commonly uh, notice. Um, uh, Leslie. One is uh, China for some reason is not counted or not measured. Okay. And second, uh, from India as well. And we've discussed this in, in the past, the submission uh, of projects into GitHub is, is actually you know, what goes on to people's resumes. Okay. And, and Indians love you know, changing jobs and as soon as a new technology revolution comes in. Okay. We've been seeing this for the last 20 odd years.

Speaker A: Okay.

Speaker B: So the, one of the drivers of people submitting their projects to Git is also to put on uh, their resumes for, for a better job change. So I'm not taking anything away from uh, from that, but that's a significant driver amongst Indian uh, professionals.

Speaker A: It is a LinkedIn of coders.

Speaker B: Exactly, it's a LinkedIn of coders. And, and I agree with Atul, what you know what he's saying, you know, infrastructure innovation, uh, innovation and application innovation. And India I think is leading on the application and has a potential to lead on the application innovation. Um, his point on indic language is you know, at uh, at last count, at its peak I think I would have counted about 32 Indic language LLM companies of which you know only one has a ground up, you know, completely different architecture. Every, everything else is a wrapper on, on a chatgpt or a llama like he mentioned. Right. And I completely understand if you know, we've been talking about this enough number of times and I've been speaking about this on conferences as well. If there is one area where India can actually innovate at a foundational level, that's probably in the non English uh LLMs, uh mode but that too that innovation needs to come ground up with a new architecture and it cannot be a wrapper on the existing uh, LLMs like the llamas or, or the genius.

Speaker A: It's going to be a challenge. It's going to be a challenge. Yeah.

Speaker B: And it requires like to Atul's point and to your point it requires the ecosystem to come in. It requires number one the data to be available.

Speaker A: Okay.

Speaker B: The digitized uh, data in Indic languages to be available, the patient capital to be available, the resources to be available and more importantly the market. Okay. Because we've seen some of these digital uh, um, you know uh, Indic languages companies. Okay. After developing all these technologies, not getting enough roi.

Speaker A: You know I always feel that it's patient capital is the main uh issue I think because you know uh, if you look at uh, the work that we have done in Indic is just look at cdac, the amount of work that it did. So you know sometimes people forget that CDAC has been the originator of most of these things. Like today you talk about this Bhashani, etc. But everything, you know all the CDAC data has been uh, subsumed by Bahashni and Bashni itself has gone into the question.

Speaker B: So we need to get back.

Speaker A: So I think Atul wants to say something language.

Speaker C: For example if you remember there's a time where Flipkart tried to convert the whole app in Hindi and various app tried that.

Speaker B: Right.

Speaker C: But that didn't work. So it doesn't. So we want to develop an index LLM and all. But who is the user? Who's going to pay for it? That's the fundamental. Is there anyone who's going to pay for it? And that's what investor want to see and that's what the whole ecosystem want to see. Apart from the government. I don't find a uh, a user who is going to pay for it. Right. So that is there are a couple

Speaker B: of, there are a couple of industries Atul to your point, um, banking. Okay. Uh and to some extent retail. But the point is is there enough ROI for the amount of.

Speaker A: Because it will be commoditized after some point.

Speaker B: Exactly.

Speaker A: See because what we are doing currently we're just using the you know the uh, chatbots they make in rule based chatbots to you know sort of LLM based chatbots and now agentic chatbots. That's what we're doing. So you know that's, I think that's what Atul is saying. Where, where, who pays for that?

Speaker B: I think we, we are discussing from two different lens. One from the lens of uh, a one from the lens of a researcher and innovator. From a, from a lens of a researcher and innovator. There is a potential for India to innovate at a foundational level in non English LLMs. But another lens is that of an entrepreneur and a businessman. Okay. Is there an roi? Okay. And both these not necessarily, you know, uh, are aligned at this point of time.

Speaker C: So we are a consumer centric market. We have so huge population and per capita income is not that high.

Speaker A: Right.

Speaker C: So like China or maybe America. So those guys have that leeway to do the experimentation with their architecture, ecosystem. Everything we do, we are everyone want to become Karoorpati next day.

Speaker A: Right.

Speaker C: So we have to have something like where you can make money with whatever the innovation you are trying to do. So a quick innovation kind of a country. UPI was a quick innovation, right? You just QR code was launched within a year and scaled. That's all. So this is something India wants. So that is why I always say if you want to want to become very successful AI company in India. So far I haven't found anyone when we are also trying to actually achieve that. You have to first grab the market, generate some revenue through your application layer. Once you start making money, then start doing the foundational research. This is how uh, American E commerce, uh, become big. Amazon started like selling some books and everything. And then once start making money then they start developing the cloud and different other businesses. Right. AWS and all right. I think this is going to come for India application layer companies will start doing making the business and then they start bringing some really new innovation. Not the copy of some foundational layer which is like LLM or something in India, but something very new which is very different than any statistical model which are being developed. So I think capital through the revenue will be a powerful push for any big innovation in India. And I think the new generation of the entrepreneur which I see after 201718 who are in this market of AI, they are trying to do that. And I believe this is something which will change the world. Uh, now coming to the chatbot walla thing also, right. I don't know how many of you have used the AI chatbot, right? I really feel bad sometime what happen I need an help and actually is responding something else though. It's a. We want to have a human interaction when you're trying to do a service, right? If I am having some problem in let's say in my wi FI connection at Airtel or JIO or for anyone I want to talk to a human that this is a problem resolved it. I do not want to have a chat with the chatbot. And this is where I still don't know what is a success rate or satisfaction rate of the people who are Using chatbot to solve the problem. I mean, I still don't have that data and I want to know what that data is.

Speaker A: No, I'm so glad you pointed that out because just recently I won't mention the company because the, but you know, I was interacting. I was forced to be in uh, interact with an AI chatbot and I kept on saying, please connect me with a human. You cannot. No, no, I can solve any problem. Look at the arrogance of that. I mean I know I'm being anthropomorphic over here and attributing human uh, things, but I was like, it's such an arrogant AI chatbot which says that I can solve all your problems even your humans can't solve. Literally said that. The only reason I didn't put that, uh, put that up on Twitter is just because I didn't want to be that mean and I realized what was happening. But uh, honestly the work does not get solved and it's really irritating.

Speaker B: Here's the thing, uh, Leslie and Atul chatbots have existed for more than 1012 years now. Earlier, for the longest of time it was rule based chatbots.

Speaker A: Okay?

Speaker B: To Atul's point, the gen AI chatbots were supposed to instill that human, human level interaction which for some reason 9 out of 10 chatbots in India don't offer. That gen was supposed to essentially bring in that human level engagement amongst chatbots. And that's why you know, your chatbot, uh, Leslie that you interacted with was so arrogant saying that I can solve it better than humans. But gen based chatbots was supposed to bring in that human level interaction, uh, clubbed with the rule based uh, problem solving. So rule based problem solving with Genai human level interaction was supposed to be the mix but you know, they are not delivered, they're not designed and delivered properly.

Speaker A: I know, I guess they will get there someday. They'll get there. But anyway, um, Atul, um, uh, now a little about your company. I think what uh, impressed me most of the company is that you know, typically since we have been talking about R and D and innovation etc, I think the more important thing for many companies in India is to create IP and you have two patented technologies that have been used in developing Jarvis and that's pretty impressive. Uh, so probably you can tell us more about you know, this, uh, just a little bit, uh, educate our readers on how you started, uh, uh, stuck U and how did you get to Jarvis very quickly.

Speaker C: So basically of course when uh, we started, our idea is to target the other data point which is called the video and image data because there are three forms, like form of the data, image video is one, text is one and the second uh, and the uh, Audio is a third one.

Speaker B: Right.

Speaker C: So most of the AI company, even the ChatGPT, everyone is focusing on the textual data because that is what the nlp, they automated and everything. Our idea from the first day was to target this image and video and that is where we found our uh, this CCTV or video data as a very powerful tool because that is where human understand that data. But to understand that data you have to look to that data. There's no automation as such. And whenever we start in 2015, and I'm talking about 2015, right, that is point you don't even know if you have a photograph somewhere in your computer. And you have to search, you have to go in each and every folder, right to where is Atul's photo. You can't type it and search it. Like today in your Apple phone you have that, you can type Atul and it will search all the tools photo, right? So that is a very uh, basic problem in the image data. Until it's not tagged you can't search it. And that is where we started with the idea was to target first the CCTV again because um, it is the largest data ah, uh, which has been created and one of the powerful devices creating it. If you send some, let's say a rocket or something to a mars, you click photograph first because human understand the photographs, right? So in that way camera is everywhere and the camera is kept as a dumb device unless until someone is not seeing it, no one knows what happened. So in that way we decided to solve the problem statement because there are two things, one is a technology and one is a what problem statement you are going to solve with that technology. For us we found out security as a big problem statement for India. Like you have 100 murders and 300 IPC crimes in every hour which is huge number for any, any, any developed country or developing country, right? So um, with that of course the security, when security comes into the mind, we started with the police and we started working with nine state police forces. Of course we work with now 11 police forces. Now we're also trying to do something with Interpol. So um, this is something we found our initial market in the CCTV camera analytics where you are using any camera without any hardware integration, anything. You just have an Internet start using Jarvis on the top of that camera. Soon we realize that camera is just not a security device. It can also Generate a lot of powerful insights which human uh, cannot collect from any other uh, data pointers. Right. For example retail industry, how many people are coming inside the store and what are the areas they are visiting inside the store. Right. Let's say I am a uh, Raymond or I am a maneuver. I am having different uh, suits and different sections. If people are going and trying the product, no one knows. You have just one data as a sales data.

Speaker A: Right?

Speaker C: Sales data do not give all the uh, detail.

Speaker A: Right.

Speaker C: So that is where we generated a non intrusive manner. We are not doing any FR or something. It's just like the human body is being identified and based on that how much time that human body is in that area you are generating different kind of uh, uh analysis on the top.

Speaker A: Right.

Speaker C: So this is something which we actually, I mean uh, innovated uh from the security to the insight level generating analytics and insight. And that is where the JARVIS was being built. Utilizing the camera to generate what made

Speaker A: you call it JARVIS inspired by Ironman.

Speaker C: Um, yes, we are a Marvel fan so I can't hide that fact that it came suddenly in the mind. But of course Jarvis, when we. So what happened? It's a very very um, uh, very uh, uh interesting story. So we were working with Up Jail, right? So that's the first time we were launching jarvis and we started with the government. I believe the uh, all the successful project in the world has been started with the government, whether it's Internet itself. Right. So we went with the uk, that

Speaker A: was with the defense.

Speaker C: So we went with the law enforcement. So we launched this JARVIS inside the jail. So initially I was writing the name Jail AI Research. So J A R came right. So then I thought let's say Vis to Lagani Paraga it will become jarvis.

Speaker A: Right.

Speaker C: So that was the way we started. Then we realized that Jail I will not be able to democratize it for other sector. So we kept it joint AI research for visual instances and stream. So that is our definition of jarvis. But yes, JARVIS comes from the Ironman and a big fan of Ironman. If you come to our office, every uh room being named with a different kind of a Marvel uh, thing. And we now go to our product. We have different modules Vision Infinity, different Marvel name in that way. So we have a lot of Marvel fan developers and of course founders are also Marvel fans. So that is from your Javi scheme

Speaker A: and excellent, excellent Jent.

Speaker B: Yeah uh, Atul. Very interesting. And uh, you know to find an Indian company which is working on um, Vision AI.

Speaker A: Okay.

Speaker B: And not just working but, but scale. I mean you guys have started in 2015, so it's almost 10 years now. Uh, talk us through some of the uh, innovation trajectory. Okay. In computer vision AI. I mean you know, all along for the last 10, 12 years, ever since AI got, got to mainstream in um, you know around 2016, it's primarily natural language processing or understanding primarily text based. Right. You know, vision based, uh, innovation was far and few in between. And from my understanding most of it came from China or at least you know, um, Chinese, uh, PhD students in American universities. Right. And being an Indian company innovating um, on vision, uh, taking the vision AI research and productizing it for different industries. Um, talk to me about um, the vision innovation journey and how you guys productized it.

Speaker A: Yeah.

Speaker C: So basically when we started the idea was to also create our own IP. We have 25 research papers, so unpublished ICCV, CVPR or DICTA. All these conferences. This year we did two research papers in CBPR and that is where the idea was to develop our own ip. Of course IP so far is called the patent filing. But we have our own research papers also and we do papers in both CV M and audio analysis. I think so far we have one of the best speaker identification technology that works on 16khz audio. So with any language you speak, it will identify who the person is speaking uh, nearby the camera. Most of the cameras stream at 16khz mic, if they have the mic. So that was the idea to basically create India's own AI technology which is having its own IP and at the same time its own market. Of course India market is quite slow. But after the chat GPT, we got that growth. Like people now trust on AI. Before that when we say someone like yeah, use uh, this in your store or use it in the police. There were a lot of of course doubt and uh, uh, they were skeptical about it. Right now you will see that uh, that growth is now fast. We have all the big players of India in the retail industry, all the, the top police forces. UP itself is the largest population in the country. They are using throughout like all our AI tools, whether it's for facial recognition. So we started with the facial recognition. I only chosen a very tough state, Punjab, where you have beards on every face, you have like different looks and all. So for the time when we launched the first facial recognition in the country with the Punjab police, and in the first month we solved around 300 cases in the pilot phase. Because when he proposed at that point of time there Was ADG Intelligence Gupta sir, who was the chief of NIA national uh, uh, this investigation agency.

Speaker B: Right.

Speaker C: Uh, he was a Punjab's dgp. And so he said yeah, it looked like very much a Hollywood script can pilot first and then we'll basically buy it. So we decided to do that and we identified 300 criminals and then it got throughout the Punjab. They call it pious. Then it went to uh, up then Bihar, uh, Haryana, Chhattisgarh, Telangana. Everyone is now using it. That was how we started. We call it person of interest or criminal of interest. And performing facial ignition on the top. Then we started adding the layer on the top. Like the violence detection, intrusion detection, crowd analysis. We did Ram Jan Bhoomi inauguration. Uh, we did summit. So that is something where uh these layers of the security and safety modules started getting added on the top. And that is where we are like you introduce. Petrochemical companies are now using for fire detection Adani Power or Haldia Power. All those guys kind uh, of countries. Right. Tata project is working with the seven of the Smart City in Rajasthan uh to monitor different kind of analysis. So that layer we developed we got a good traction because we are doing it for native India. We are developing that model. We understand two megapixel camera. If you talk about China or Japan, they have 5 megapixel camera, 10 megapixel camera. M not that crowded. Clear field of view. For us it is like Indian Nepal border. If you see the crowd, you will not even understand anything what is happening in front of the camera.

Speaker B: Visual noise. Visual noise and low, low quality.

Speaker C: Uh, so that is where we developed a lot of our modules which became very uh, important for the police forces. We are working with them for the last six, seven years. And still there are they being integrated

Speaker A: with drones and with other kind of things.

Speaker C: We started doing with the private sector. First drones. Haldiya Petrochemical is the one which is using now we are doing with Mirjapur police. Mirjapur uh, is a series but uh. So Mirjapur has a temple called Vindhachal which is one of the saktipeet of this. So they are using for the crowd monitoring Naratra. They have a lot of crowd there.

Speaker A: Right.

Speaker C: So we did that and we are also doing couple of um, of course uh, intelligence project on the drones with the various uh, special task forces of the country.

Speaker A: Yeah, Atul, typically you know, when we talk about monitoring there's always that negative way of looking at it. You know. Because on one one hand of course clearly when we talk about surveillance it's all about Safet companies will say this is all about your safety. Just like how you know that, uh, skepticism, employer versus employee. Like, okay, in a sense, if I'm using a monitoring tool, what are you monitoring me for? So there's this whole question of, uh, you know, privacy, which is very, very important. I'm sure you take a lot of care about that. But it will be nice if you can educate the readers also as to what are the, you know, kind of safeguards that are being built in so that privacy is not being violated under uh, you know, unless it's under those exceptional, uh, you know, circumstances where national sovereignty or something is being compromised.

Speaker C: Exactly. So there are two components. One is a security part and one is a retail insights and all those stuff right in the security module. I believe India need. There's a fine line of always in between security and privacy. Because when you have security, then of course you have the access of certain cameras and all. But if you talk about CCTV camera, when you see a CCTV camera, you feel okay, someone is watching. But when you have this thing in your hand, right. You don't know who is monitoring you then what CCTV is. But anyway, in our case, when we do our video analytics, whether for the police or whether it's for the private sector, we not deploy Jarvis on our own cloud. We do not actually own a cloud. Right?

Speaker B: Uh-huh.

Speaker C: The company who are making money. Because when we try to beat, let's say uh, privacy, we want to make money. And how we'll make the money, we'll start having the data, uh, like everything. Right. We're not doing that. That we are trying to become an apple of this world. We will deploy our own cloud. We will be not doing any facial equation for any retail industry. I mean that's on record. I'm saying it. We are not doing any FR. Or any identification re identification technology. And that is why we call it GDPR compliance. We are being audited every month by a GDPR auditors.

Speaker B: Right.

Speaker C: In the government sector, we are deploying in the government data center. No data is going on cloud. That means the, the sovereignty of the whole data has to be there. Because let's say Google will not work in India. So it should not like if we are technology clues.

Speaker A: Right.

Speaker C: So this is something which is where we are um, uh, maintaining all those standard which are important, which is a SoC2, which is American standard for data privacy or GDPR, which is European standard. And I think India is also coming with their own data protection law. So we are already uh, following certain and other similar organizations thing and in the private sector we are not doing any FR or something.

Speaker A: So there we have the DPDP rules.

Speaker C: Yeah. And the most important part, having said all these pointers. Right. The country which is getting a five year old girl getting raped somewhere. Right. Deserves some kind of a surveillance. We are not Finland, we are not Ireland, we are not Scotland. We are a country which is doing 100 murders in every hour. Uh, 300 IPC crimes in every hour. So there has to be some tech which if you do a crime you have to be identified. Right. Uh, and that is where you see CCTV cameras are solving lot of cases in the India. I will just go to the newspaper. You'll find out there was CCTV footage. A famous case in Patna recently. A businessman got killed. It was the main news ah. Somewhere. Right. So all these things being solved by the cct.

Speaker A: Uh, this I'm not uh, you know, digressing but there's a very important when, when we study law, um, there's a very important principle in law that you allow uh, you know, one, uh, hundred people to uh, you know it's okay if you allow 100 people to escape the criminal justice system but you will not allow one innocent man to be convicted. Okay. Now what happens in the case of AI it hallucinates. It can create a lot of errors. Can this principle be violated?

Speaker B: No.

Speaker C: So basically what happened in our case. So there are two things. Are you giving the whole control to the AI? No, it is a semi controlled environment. We are which is able. Which is helping you to identify a kind of. So we are. When you're performing let's say facial criminals, it's just on the face data we are taking the audio data is another mode of data which is also trying to match with the audio os. Let's say you found a person who's being suspicious person in some roadside checking.

Speaker B: Right.

Speaker C: So you're also taking the audio sample. Then you're also calling to the respective police station with the details. It doesn't mean it's completely automated like China where you just found out something and then you uh implicated that person that this guy is the one who did human.

Speaker A: Supervisor.

Speaker C: Supervisor. We're not giving that control. And in all these things all and every activity of human or police is being monitored. There's a record of it. So if a data, for example some police clicked my photo to add into the database, it is not directly going to the database. It has to be approved by the DCP or The SP of the District through the otp. That means the OTP has been verifying that this guy is a criminal and it's going on record. So SCP sp, Superintendent of Police or DCP is actually finding him as a crime. So he's also staying on record. Right. Of course it goes through a D.C. where we district Crime Record Bureau. After that data get added and any verification and everything goes through that. So we have lot of check and balance on that. If you go through our different kind of a tool, of course we have given that access to BBC or uh, different. Similar kind of organization who write quite um, very straight towards the government. Right. So we given that access to those guys and we give it them like go and do all your investigation and not find any such cases against us. Where we have given that kind of freedom where the police can exploit us because. And then we are also a common citizen. That technology can also be utilized against. I'm not the police.

Speaker A: Absolutely.

Speaker B: Exactly, exactly.

Speaker C: That check balance. And that is, I'm m saying on record like we haven't given that freedom event and some of the cases we didn't even help the government. For example, the former protest and I'm again seeing on record, they asked us to help us, we said no. Yeah, this is their protest and this is a fundamental rights for them. We can't help you. There were certain cases where it was attack on the embassies or something. So we did that. So we take pride. We also take that responsibility.

Speaker A: Not going to make money at any cost.

Speaker B: So when he said we want to be the apple of this industry, that's what he had in mind. I mean we will uh, retain the fundamental rights of citizens. Atul, I have a question which is a follow up to what you said and preclude to the question that Leslie asked. It's somewhere in between. Uh, you mentioned that uh, anyone who's committed a crime needs to be identified and punished. And that's where your uh, solution helps. But the power of AI and the power of data and digital data and pattern recognition, is that going forward, uh, it can be predicted as well. I mean the famous Minority Report movie, uh, right now you guys are identifying, okay, uh, the criminals or the people, uh, who have committed the crime. But uh, the power of the technology is that, you know, it can also to a fairly accurate uh, extent predict. Okay, number one, I have two parts of the question. You know, number one is that thought process for a solution being thought through. And second is that request coming in from the police and the vigilant uh, services to require that. And three, will we ever see that kind of a technology? I mean people keep saying minority report is going to be a future very soon.

Speaker A: Looks like you want to see it.

Speaker B: No, in some parts of the world, I mean considering what they're doing.

Speaker C: So this is a very typical example of drugs things which is being coming from uh, Pakistan through Punjab through Delhi to uh, different things. Right?

Speaker B: So right.

Speaker C: These networks get activated in different uh, weather time different uh, year time, different uh, those kind of a time. Of course when you collect the data of criminals, you also also collect those information fir report that you have a background why this person is a criminal.

Speaker A: Right?

Speaker C: So for our person of interest module, which is a part of Jarvis facial recognition, you have that data. But that data also give you a lot of patterns. Like for example M. These are the areas which become the hot spots of the crime because people use this trade route to basically bring the drugs. Because drugs is a business for them. It's not a um, uh, material of something. Uh, this thing it is more about like for example in Afghanistan someone is selling drugs of 1 kg in 1 lakh rupees. In let's say in Pakistan it's 10 kg 10 lakh rupees. At border it's 1 crore rupees. And in Delhi it's a 5 crore rupees. The 1 lakh 5 crore rupees in Delhi that's a business.

Speaker B: Right?

Speaker C: So you have to identify those patterns which is coming through that data if someone is being caught. Right? So yes, we are doing that kind of a predictive analysis now we have started because very important part for any predictive analysis is just not a data, uh, the continuous stream of data is important. Exactly right. So that you can. And the most important part again in Indian crime thing is like 70% of our criminals are repetitive come out again do the crime. They again do the. Because Indian judiciary like you said, right? So that is something which is also giving a very good benefit of doubt as they say out to the criminals, right? And that is where they are just exploiting the system. That is where the predictive policing is really important for a country like us, where we have a 200 police person for 1 lakh people, which is one of the lowest in the world. And the crime is a super super, super high. Right? So that is something which is where without AI, without automation it is impossible. Without technology it is impossible.

Speaker A: So folks, I uh, think very uh, clearly the ground rules have been established that AI and automation is required. Uh, you know if you want to have a proper surveillance system or proper guard, you know. But a lot of guardrails are also being built in. There's a. I think the comforting factor, as per what Atul has also informed us and, you know, clearly, uh, sort of elaborated for us how humans are in the loop, basically. And it's an extremely important thing. Yes, there is always that, you know, concern that it might turn out to be some kind of a minority report, as Jayant rightly said. And it's legitimate. It's as good as asking whether AI will at some point in time, you know, become AGI or become even artificial super intelligence. We do ask those questions. Now. We know that it's easier said than done because, uh. Uh. It's, uh. You know, not as. Because there are plenty of challenges that, uh. A, uh. Company like Starkey also faces in implementing these, uh, technologies. Um, I know, uh, for lack of time, we have not had, uh, you know, not been able to discuss many of these aspects.

Speaker B: But I'm sure we'll be able to

Speaker A: do it, uh, sometime some other time, as they say. For now, of course, folks, uh. Uh. I would like to thank Atul for his, uh, time and, you know, explanations. Very serious kind of, you know, detailed explanations. Uh, and I think, uh, many people, many of us would be enlightened as to how AI is being used in US Vision. AI and of course, the whole idea of how, uh, India is building its whole AI ecosystem. Once again, thank you for your time and, uh, Jayant, you two also have a lovely weekend. And hopefully you'll be back to your effervescent, uh, self. Bye. Next week. Okay.

Speaker B: That was very informative. Thank you, folks.

Speaker A: You also have a lovely weekend.

Speaker C: Thank you. Thanks.

Speaker A: Bye.