Bengaluru Startup’s Bold Vision: How Their AI Can Improve Public Safety, Shape Key Industries, From Aviation To Defence

In his address to the nation on Independence Day 2024, Prime Minister Narendra Modi set a possibly surprising, and since then probably forgotten, lakshya (goal) for India: quality.

He said it was important for India to be recognised for its quality and standards. "We must strive to make Indian standards synonymous with international standards," he said, because that would help Indian products find greater acceptance in the global market.

Though this call might seem innocuous in the grand scheme of India's things, it's actually a key driver for India's march towards becoming the world's third-largest economy.

This writer recently visited the office of Camcom Technologies, an artificial intelligence (AI) company based in Bengaluru, Karnataka, which has made quality its primary business.

Among other things, CamCom has pioneered the world's first large vision model (LVM) — different from the large language model (LLM) we are so accustomed to hearing about — for defect detection and assessment in a variety of products across sectors.

The LVM is an advanced AI system designed to understand and process visual data. It turns out that this new technology does not just elevate quality in the manufacturing sector, but also has a tangible public safety impact, as evidenced by its successful work in Saudi Arabia.

Now, Ajith Nayar, the company's co-founder and chief executive officer (CEO), is speaking to government officials in India and around the world, offering his technology as a way to improve the quality of life of people everywhere.

This writer met Nayar at his Bengaluru office for a conversation that exceeded well over an hour. Here are the edited excerpts:

Ajith, your company, CamCom Technologies, was recently recognised by NASSCOM as an AI gamechanger in the manufacturing category. Could you give us a sense of the work that's earned you this and other accolades in recent years?

We are in the business of quality. That quality could be manufacturing quality, service quality, or life quality. So, this (the NASSCOM recognition) is specific to manufacturing quality.

In products, the asset that is being manufactured tends to have defects as a part of the process. It could be anything from an issue with the paint job to a dent in the case of an automobile, a glass broken in the case of a mobile phone, or even something like a scratch on the surface of any manufactured product.

It's been evident that India is not synonymous with quality, as is Germany or Japan. So, one of the things that we at CamCom wanted to do was to make India synonymous with quality. That is actually CamCom's underlying mission.

Now, quality in manufacturing is a very human-intensive job. Don't get me wrong — the human eyes are still the world's best cameras. No camera can beat the human eye. But we have an extremely subjective processor, which is our brain. There is nothing wrong with that because that puts us on top of the food pyramid, if I may put it that way. Because we come to decisions based on a balance of consideration. But quality is not a balance-of-consideration decision. It's a binary. The answer is only yes or no on a defined set of parameters. There is no maybe, possibly, plausibly, or justifiably.

When you put humans into something like that and tell them to track a 50-micron defect, the human eye cannot see a 50-micron defect. It would help if you had specialised equipment for that. And with humans, it's essentially diminishing returns from the time you start to the end of your shift. What you need there is objectivity, and you need consistency in reporting. Because anything that you find in terms of a defect, if it's not caught at the station where it should be caught and goes downstream, god forbid, it enters the market, there's a huge recall cost.

Now, about products: 63 percent of the world's products are made from metal, plastic, or glass. Quality impacts the three most important fundamentals in business: customer satisfaction, top line, and bottom line. 

We won this award specifically for automotive manufacturing, which we had done for one of the largest SUV (sports utility vehicle) manufacturers in India. We caught 55 percent more defects than the human eye. And we do that not just for them. We also do it for the second-largest bike manufacturer in India, where we look at the fuel tanks and many others worldwide.

The Government of India has announced a policy of 'zero defect, zero effect' (ZED). If I may, we are the torchbearers, at the risk of sounding immodest. The ability to marry technology into industry 4.0 standards is a basic bedrock that the Prime Minister (Narendra Modi) articulated. And that's precisely what we do.

It's interesting you say that because when I was listening to Prime Minister Narendra Modi's Independence Day address, he specifically mentioned that it was his aspiration to make sure Indian standards are aligned with global benchmarks. And because we had already planned this conversation, I thought your work was in essence very much in that direction. Could you speak to the serious need for quality assurance, quality standards in India? 

Actually, let me flip it the other way around. I don't understand why globally they cannot follow Indian standards. I think we should redefine international standards and set the gold standard for ourselves and for the world to emulate.

If you look at Japan, how did this whole concept of "kaizen," you know, Six Sigma, Lean, etc, actually come about? It comes from the ability to create a process and continuously improve the process. Once done, repeatability ensures quality. Now, repeatability is the one thing the human mind does not particularly like, but you need to bring in those SOPs (standard operating procedures) to ensure that the repeatability happens continuously. That is the whole idea behind an assembly line, as Henry Ford learned from the meat district of New York.

Even today, if you go to the Toyota, Tesla, Daimler, Volkswagen, and Bentley factories in the UK (United Kingdom), US (United States), Japan, or India, and my colleagues recently did, they came back and said they were surprised at how labour-intensive the operations were. Visual inspections have always been labour-intensive.

It is not that the camera didn't exist or that people didn't know that it would probably be able to do it more effectively and consistently. It is that, one, there is this fear of people losing jobs, which has always been there. The second part of that is that it’s cheaper. 

If you look at an ROI (return on investment) calculation that most people do in India with regard to inspection time or contract employees, the first thing they say is, 'Oh, I can have 20 people instead of the camera'. Twenty people doing what? At the very end, think about whether they are doing it well. Human beings are extremely good at doing certain things. They are bad at doing other things. Because subjectivity is ingrained in us, we are different from the other species in the animal kingdom.

Objectivity is something that, when it is asked for, when it is called for, we falter. Because light, the time of day, whether you had a heavy meal, whether it was biryani or pizza, all those things come into play. Did you have a fight with your wife and wake up on the wrong side of the bed? All these things come into play. So, instead of that, we start looking at creating and charting new SOPs.

This is something that, when I had the opportunity to meet with Subrahmanyam sir, the CEO of NITI Aayog, and also I was fortunate enough to meet the Prime Minister sometime back when I was briefing him on LVMs (large vision models), I said, ‘Let us define our standards first.’

Industry 4.0 as a standard has been, of course, rolled out across the world. But in that also, the technology is a recommendation. Instead, let us make sure that it happens. And if we are able to link the ZED programme, the 'zero defect, zero effect' programme, with the PLI (production-linked incentive scheme), all of this will fall into play. 

If the Government of India and the Prime Minister specifically want India to become the third-largest economy in the world, it is built on one thing, and that is exports. Exports have no meaning if they are of poor quality. So, that is what I feel that the government should focus on.

You mentioned technology and also briefly something that's very important to your business, the LVM, or the large vision model. Of course, AI models are all the rage right now, but they're specifically the large language models (LLMs). You've innovated what's called a ‘large vision model’. Just for the benefit of our readers, could you talk to me about the nuts and bolts of this large vision model and how it's different from the generative AI space?

The fundamental difference between LLMs and LVMs is words versus visuals. What is the difference between a word and a picture? The picture is worth a thousand words (laughs). So, you can ingest as many as possible and then get an output.

Now while LLMs are all the talk, the ubiquitousness of LLMs today has led to a situation where plagiarism is on the rise. Because it has ingested literally everything in the public domain, not particularly heeding copyrights or patents.

LLMs have to find use cases. It is one thing for LLMs to act as chatbots or for LLMs to provide you essentially, let's say, a predefined set of essays, for example, based on the input that you give. But somebody has to get the LLM to become more domain-specific. Because the way that a receptionist or a BPO sector worker would speak and interact is very different from, let's say, a car service mechanic. So, there has to be a certain depth of domain that needs to be included in LLMs.

Now, LVMs cannot be generic. They have to be focused. So, in our case, we have created a deep learning vision model that is specifically focused on defect and damage — defect during production and damage in the aftermarket. 

At the same time, it is so easily transferable across multiple industry segments, and CamCom is horizontal rather than vertical because of this. Defects and damage are in everything. A defect on any surface is a defect. There is a tear in metal and a tear in plastic; glass cannot have a tear, but it can be broken. At the end of the day, it's a tear. Now, the system has to learn, and that is what our system has learned over a period of time. Defect or damage detection is form-agnostic and function-agnostic.

The moment it (the system) can understand that a glass is broken, it's essentially just mapping it to that and saying, okay, this is the problem, this is the severity of the problem. Of course, the defect needs to be immediately notified. The typical throughput time is just about 10 to 15 seconds because the vehicle is constantly moving, or anything that's on an assembly line is constantly moving. So, all the computers are at the edge, and we can provide that, so immediate rectification can happen.

And, of course, there's an associated root cause analysis. So, for example, if the system catches, let's say, what’s called an "orange peel," which is a drip of paint. If you find it on five vehicles, what that means is not that there is an orange peel. This means that the humidity in your paint box is high. So, fix that first. If there is too much dust in the paint, the AQI in the paint box has gone for a toss. So, these are things that enable them to quickly fix it. Otherwise, you run another 20 vehicles through that. Painting is a very expensive proposition!

So, these are things that a vision model enables because it is essentially learning from the images that it is seeing. Once it has learned it, the ability for it to incrementally increase its accuracy is very high. We also happen to be in a good position where our customers are sending us curated data.

Customers will do a review because, at the end of the day, they also want to know if this is better than what the human is catching. And when they do the review, typically in the case of defect assessment, we are at 99.97 per cent. It cannot go beyond that. That's also because we control the input. The input is not some random image; it is our hardware, a retrofit bespoke thing that sits on top of the existing assembly line and is designed to capture the best input possible for the AI platform to provide the best results.

But in the case of aftermarket damage, pictures come to us from a mobile device with a camera. Any mobile device with a camera built after 2017 can capture images in 4K resolution. That’s good enough for the system. So, when the likes of an HDFC Ergo, SBI General, Bajaj Allianz, when they underwrite a car or settle a claim, and somebody is clicking random pictures from, let's say, somewhere in Hoshiarpur, and just using their phones, we have the ability today to say that this particular claim can be settled right now. Because IRDA says everything under Rs 50,000 does not require the services of a loss assessor or surveyor.

These images are good enough because the system can tell them with 96 per cent accuracy that this is what it is. That automatically increases the productivity of the various surveyors on the roads, reduces the overhead costs otherwise associated, and has a positive impact on the claims ratio.

Recently, the banking secretary said that they were asking public-sector general insurance companies like National, Oriental, and United to get out of the motor business because their claims ratio was abysmal. But seven of India's top 10 private-sector insurance companies use our solution precisely for that.

The Government of India has already made significant strides in harnessing technology to enable scale, ease of use, and easy access to services for our 140-crore population. UPI (Unified Payments Interface) is a classic example, and now the move to satellite-based tolling systems. With people like the Honourable PM at the helm of affairs, who is himself very, very tech-savvy, and complemented by stalwarts including Rajnath Singh ji, Nirmala Sitharaman ji, Nitin Gadkari ji, Ashwini Vaishnaw ji, the pace of adoption is increasing.

Can you talk about some of the specific initiatives or projects your company is undertaking? 

I think one of the biggest, and one that we are probably the most boastful about, is public safety. Three years ago, the Kingdom of Saudi Arabia — MBS (Crown Prince Mohammed bin Salman), as you know, a person who is a visionary, gave us a breakthrough. 2030, Riyadh and other cities in Saudi Arabia will be world-class. Obviously, with the PIF (Saudi Arabia's sovereign Public Investment Fund) having $170 billion, it was not that difficult to decide.

Riyadh, their capital city, is in the middle of a desert. Now, that (Riyadh) is what he wants to create as a showcase. The infrastructure over some time had crumbled. 38 square kilometres in the city are currently demolished. And a brand new CBD (central business district) is coming up. And all of that is going to happen before 2030. So, he just woke up one fine day and said, 'So be it, khul ja sim sim' (laughs).

This falls under the Ministry of Municipal and Rural Affairs and Housing. So, since MBS said this had to happen, they started looking at how to do it. Firstly, you need to understand what the issue is. The issues have to be reported. Then, they can go ahead and rectify it.

They issued the citizens with an app called 'Snap and Send'. The citizens were delighted. Two and a half million incidents were reported, and they (the government) didn't know what to do with it. So, much like any government, they set up a BPO. What will a BPO do looking at a picture? There's no way to derive metadata from it.

When this (proposal) initially came to us, and I am being 100 per cent honest with you, I remember I was in Dubai at that time. I said, 'Look, this is not something we do.' After a request to try it out, I said okay. And I remember telling a colleague, ‘No changes on the platform. Just run it through the engine and tell me the output.’

About three or four days after that, I came to India. Then somebody showed me the output: it was 82.5 per cent accurate. I honestly fell out of my chair. I thought there was something wrong with my engine. So we went back and looked at the input data, and that is when I realised what he had done inadvertently was paying off. Metal, plastic, glass.

A manhole cover missing is a piece of metal missing as far as the system is concerned. Construction debris will always have metal, plastic, and glass. And then the transition became that much easier. Now, we have added asphalt, concrete, rubber, and a bunch of other things.

We contracted with the Saudi government in April or May (2023). I think we went live with them last November. In July (2024), they contracted with us for 4 million inspections per annum. They've already crossed a crore of inspections!

At the G20 event, I was one of the speakers at Startup20. And a gentleman was talking about UN SDGs (United Nations Sustainable Development Goals). I sat there, and I listened, and when he came to point number 11, he said, 'Citizen participation and urban planning', 'Conservation of natural and manmade heritage sites', and 'Waste segregation and management' are the key implementation requirements for compliance. It suddenly made a lot of sense. Inadvertently, we ended up helping the government comply with UN SDG 11. 

Saudi Arabia became the first country in the world to use AI to achieve this, and we, the first company in the world to enable a country to become compliant using AI.

You speak about the Saudi Arabia experiment. Honestly, Indian cities absolutely need that. Have you had any takers from India? 

The meeting with the Hon Prime Minister happened because of what we did in Saudi Arabia. 

He opined that this should definitely be implemented in all the smart cities in India and that Indian technology must also be given to our closest friends and allies, as well as to the Global South and beyond those horizons. 

Dr Himanta Biswa Sarma, the Hon CM of Assam, was very keen to understand the solution, and we recently presented it to him and the team in Guwahati. They say startups inherently move with alacrity and agility. Have you seen state government machinery move that way? I saw it for the first time in Assam. The political and bureaucratic arms are coordinating perfectly to get things done. 

We are also in talks with Andhra Pradesh, Goa, Gujarat, Uttar Pradesh, and Jammu and Kashmir. During a visit to India, the President of Mauritius chanced upon a press article that talked about what we had done in Saudi Arabia. They want to become the first country in Africa to be UN SDG-compliant. We also have discussions going on with Qatar, Dubai, Abu Dhabi, Harris County in the United States, and Aguascalientes in the city of Mexico. One hundred and ninety-three countries are signatories to the UN SDGs. 2030 is the deadline for compliance, so everybody is running around trying to figure out what can be done. We are probably the fastest way to become UN SDG-11 compliant. This makes us extremely happy because we are making a distinct impact on people's quality of life.

When I asked the Saudi minister in charge what is the actual impact on the ground that they have had, he said you've reduced the number of fatalities. A manhole cover missing is a clear and present danger. And he said it (the number of falls) has essentially dropped by 60 per cent. Because somebody has reported it, and we have been able to rectify it.

Recently, I met the Indian Olympic Association chairperson, P T Usha madam, in Delhi. She was so interested in our technology that she immediately connected me to the people in Paris, saying that you (Paris) are the first sustainable Olympics in the world. Why don't you look at this technology? You'll also become UN SDG-compliant. Unfortunately, the day I landed there, President Macron dissolved the Parliament (laughs).

But we still have a lot of interest coming. The city of Dijon, the city of Cannes. The mayor of Paris, she was very keen.

Every smart city in India has a command-and-control centre. We are probably the easiest to plug in because any feed that is coming in, it really doesn't matter what the source is. You have to find a pothole, overflowing garbage, or a naked wire, for example. You will find it; we will tag it.

The other impact of that (the work in Saudi Arabia) was — and it literally had me bewildered — the Amanas, or the municipalities (in Saudi Arabia), 19 of them, increased their revenues by 51 per cent. I asked, how did that happen? Automatic penalties. Every public infrastructure has a service contract associated with it.

Now, say you have a building permit and are building something there. There is construction debris outside, which is not allowed. The person who has been issued the building permit is automatically charged. The road has a pothole. The road contractor who is supposed to upkeep the road is automatically charged.

The revenues have gone up. Of all the work that we did, this is the one that has a tangible impact at multiple levels: public safety, compliance, enhanced revenues, enhanced quality of life, and sustainability of urban areas.

We’ve looked at the civic aspect, urban planning. What other sectors do you think your tech could come in handy? Off the top of my head, one area I think about is airline accidents, for instance. Any other areas? 

You hit the nail on the head. Airlines are a very interesting space. We intend to enter this industry this year, and some discussions are already underway.

An aircraft, at the end of the day, is nothing more than metal, plastic, and glass. There is the MRO concept — maintenance, repair, overhaul. Every 20-24 months, it (an airline) has to be towed into a hangar. Getting this done takes one to four weeks and 6,000 man hours. A scaffolding needs to be set up, and then exterior and interior damage checks are done. That is called a Category C check. There are also Cat A and B checks that are not this time-consuming.

What we are proposing is that when towing the aircraft into the MRO, we will put a rig on top of the hangar and one on the floor of the hangar. We will collect the pictures as the aircraft is being towed. You need to put up scaffolding and focus only on the place where there is exterior damage. Why even put up that scaffolding and have all these people, four engineers, walking around the aircraft? 

The most common check is, of course, the preflight check. You have the captain and/or the copilot, you have service engineers, normally three of them, under the aircraft, the belly of the aircraft. They walk around the aircraft trying to see if there is any problem and that is it. It gets signed off by the copilot or the captain without even knowing what they are signing off on. The flight takes off. Something happens mid-air. The person who signed off on it is dead.

There is no visual audit trail. Can you imagine that? There is no visual audit trail of a commercial airline with normally 180-plus people taking off. So, we were like, how can that even be? Is it because you have four people walking around anyway, so if one guy misses, the other guy will catch it? Why not a handheld device that can capture the condition of an aircraft before it takes off?

For example, this was something that came up in a conversation with a senior official in the Defence Ministry. We have underwater submarine cables across the entire Bay of Bengal, Indian Ocean, and Arabian Sea. They are the cables that we use for our submarines to communicate with us.

Now, these cables are undersea. Obviously, there will be issues. Right now, if something goes wrong, you have to send somebody to understand where the problem is first. There is no way for you to say there is a likelihood that this will fail. It's 100 per cent dependent on typical preventive maintenance schedules. Is there a way for computer vision to do something about it?

Now, underwater computer vision is probably one of the most difficult things to do because, firstly, water distorts images. But now we have come up with certain things that could potentially help us do that. But can you imagine the kind of impact something like that will have? Because the entire internet system today is underwater cables. We are talking about a $100 billion industry in that sense.

There are huge opportunities available out there, but we are a small team — we are only 44 people, full-time employees, and another 38 contractors, interns, etc. There is a bandwidth limitation. So, our immediate focus will continue to remain on the automotive, insurance, and public safety sectors.

We will add adjacent industry use cases, but to build a brand-new model, such as underwater vision, we will probably need to add multiple layers to our damage detection and assessment model. The filters and IPs (intellectual property) needed to get those images to a level where the vision can learn from them will take a little bit of time.

Broadly speaking, how do you see visual AI playing a role in India's growth story?

Oh, 100 per cent. At the end of the day, the bulk of quality inspections are visual inspections. I'll give you a simple example. Do you know what the average life of a mobile phone in India is? Seven months. People are trading in their phones and buying new phones.

Now when you are trading in a phone, there is a delivery boy who comes and looks at your phone and says everything is good. He is going off a small diagnostics engine, but how will he know what the quality of the phone is? So, they have now asked us, ‘Can you create a small contraption for us so the phone can be put in and you guys take the pictures, first and foremost, to understand what the quality of the phone is and should it have an impact on the price that is being paid?’ 

Secondly, can it (the phone) even be refurbished? Why should it even go to the refurbishment centre? Because the transportation cost associated with lugging it back to a recycling plant will be more. So, it (visual AI) is going to have an impact without a doubt.

And I am not just talking about AI from my perspective of computer vision. I'm talking from an overall perspective. See, 1886, wasn't it when Benz invented the motorcar? 1902 was when Ford essentially set up the plant in the United States. The First World War was still dependent on donkeys and horses. World War 2 was when the automobile essentially started showing its prowess.

Look at the change. At the end of World War 1, there were 100 million horse carts and probably less than 1,000 or so automobiles. At the end of World War 2, there were 100 million automobiles and zero horse cart drivers. That does not mean all horse cart drivers became automobile drivers (laughs). 

Every disruption brings opportunity. All we have to understand is that it is not about how many jobs will go. It is about how well this can be run and the impact that it's creating. Because, at the end of the day, whether you put 100 people on the job and they come up with a lousy product or you have zero people on the job and a good product, what makes more sense? Because the money for those 100 people can come from one good product.

In the short term, yes, there is going to be a significant amount of disruption. I think that the BPO industry is going to see significant disruption. I see service jobs, where the human factor is not that important, the likes of insurance surveys (going away). But they will upscale, they will upgrade, they will do something that they will actually be able to contribute rather than become yet another donkey in a line of donkeys, not knowing what they are pulling. They become parts of a wheel with a clear understanding of what that wheel is going out for.

And I think more and more jobs will be created. Elder care. Do you think technology can replace elder care? No, it would be best if you had human empathy for that (role). And there are so many such jobs. A nurse will never go away. A teacher. If you want your child to essentially learn from LLMs, all they will be doing is plagiarising. There is no original thought. Yes, people say there are very few original thoughts, but nevertheless. So, there will definitely be an impact. 

One other thing. In 2017, the UAE was the first country to appoint a minister of AI. Before the EU and the Americans woke up, they introduced a minister of AI. He set the path for every department to have a head of AI today. The government itself is changing. The same thing happened in Rwanda. Kigali has a full-time head of AI. In Eastern Europe, the Baltics, and Estonia, 100 per cent are run by artificial intelligence as far as government citizen services are concerned. Why can't we do it?

You bring us to a very important point. I imagine legislators and policy makers are really sweating right now because they are having to keep up, one, with the tech, and two, with the speed at which it is progressing. I think it's a tough gig. But how do you see them responding to this AI wave? Should they be ultra-aggressive, trying to stay ahead of the curve? Should they be cautious, taking it easy? 

I don't think there is any need to be ultra-aggressive. Because, at the end of the day, it's a hype cycle. Guys who started the disruption or who joined at the lower end of the cycle essentially tend to understand what they are building. And then, at the top of the hype cycle is when typically the regulators come in. And by that time, it (the hype) has already started going down because people understand the pitfalls associated with it.

With any new disruptive technology, regulators will have a difficult time understanding what it is because of its impact. Uber is the classic example. People did not even understand what it was. But the impact that it had was so big that they had no choice. And that essentially led to what? The person who pioneered the technology was at the high table when the policy was written. Whether the policy is good or bad, at the end of the day, even the person who has the disruptive technology doesn't know.

See, we talk about singularity insofar as artificial intelligence is concerned. In specific areas, that has already been achieved. For example, in the case of the solution that we provide for defect detection, forget singularity, we are better than the humans. We can prove that. In certain areas, that has already been achieved. And that will continue to be achieved over a period of time. But is the machine going to completely replace humans? I personally don't think so.

And at the end of the day, if that is what we are going towards, how are you going to stop it? Is policy going to stop it? No. But people today working on these technologies understand the importance of regulation. The reason why suddenly everybody and his grandfather are standing up and saying, 'There has to be regulation', is they saying, 'I have made enough money and really don’t know how to deal with it. I have let the genie out of the bottle, and now it's your problem.' Honestly, that is what is happening. 

The question is, can any policymaker have the kind of foresight to be able to take these calls right now? The answer is no. I mean, it's just not possible. Because every day, something is coming up. Certain things, I guess, we need to leave it to the future to decide. But at the same time, there are certain things that the government can 100 per cent regulate. Anything related to human data, specifically medical records.

Again, I don't think we should ape what is going on in other parts of the world. It makes no sense. You can't define certain sets of data as personal when you know that it really is not personal. And they just become unnecessarily anal about it. Because controlling the data does not give you anything other than brownie points. 

Fortunately, we have also moved away from one-size-fits-all AI to more sovereign-focused AI. And that is very important because we deal with regulated industries, such as insurance. There is insurance data that should never leave the country because there is no guarantee as to what it will be used for today.

In India, we do have a problem with hardware. And I think India should definitely focus on trying to create more fabs (chip fabrication plants) and look at the possibility of indigenously manufactured GPUs (graphics processing units). Because that is where we are going. Or even maybe something better, for example, Cerebras, the guys out of the US, which recently the Abu Dhabi sovereign fund invested in. It's a chip, which does 100 million more computations at the same time. So almost equivalent to a GPU.

I think we should look at more and more innovation coming in the hardware space for us to be able to take a lot of what's happening in the software space and keep it local. Because of the data leakage. We definitely need a proper AI policy with regard to hardware. I think more and more people are going to want that kind of compute. And I still don't know why we don't have a fab. Maybe the government should look at having four or five fabs.

Nandan Nilekani repeatedly says India should be the ‘AI use-case capital of the world’. Are you thinking in that direction, or do you prefer more foundational work?

I completely agree. Because foundational work requires significant compute. We don't have that kind of compute in India. I say we do exactly what we did with IT services. They essentially went up the curve and stayed at the curve. And they didn't know what to do with it. How do you reduce costs? At the end of the day, you need to reduce costs. That is when Indian IT services companies came in and built multiple layers of IT services. Which, of course, is now at the end of the curve. Now we don't know what to do with it (laughs). That cycle will always be there.

In India, building foundational models does not make any sense. We don't have the kind of money to invest. Instead, we should focus on creating as many use cases as possible out of that. We don't build foundational models. We essentially take a fundamental model and build IP on top of it. And let them do the hard work (laughs).

All this nonsense about no significant contribution from an intellectual property standpoint is complete bunkum. There's so much being generated in India. The problem is that people don't know it unless it immediately affects them. How many people in India know that we impact public safety? Nobody will know unless someone goes and tells them.

What's been your experience of the business climate? Ease of doing business is a big thing for the government and the country. What's it been like working as an innovator, as an entrepreneur? 

Let me put that at multiple levels. One is ease of doing business in terms of setting up a business, in terms of setting up a bank account, and all that. India is way ahead of the rest of the world. It takes me less than 48 hours to set up a company, and I have 10 banks lining up to give me a bank account. You get a current account done in less than 24 hours, and the choice is yours. So that is one thing I think the government has definitely taken care of.

Intellectual property. There are enough and more incentives and subsidies by the government, which, by the way, they have actually empanelled lawyers for it. So, when we go and speak to a lawyer, they are adding that financial incentive and passing it on to us. So, we don't end up paying as much as you would otherwise pay to file an IP. So that is again something through Startup India, Atmanirbhar Bharat, and all that, these things have been taken care of.

One thing that’s there — and that is not actually because of the government, and I honestly don't think the government should be getting into that also. The venture capital system in India is not mature, unlike what you find in Silicon Valley. So, when you essentially use a Valley playbook and apply that to a venture in India, it doesn't work.

Should the government step into that? No, I will not say that. However, the government can and should definitely help with working capital requirements. And that is something that I have seen across India. Most of the founders I speak with have a problem with working capital. And the reason for that is, firstly, in India, working capital loans are given to profitable businesses or against collateral. By its very nature, a software company firstly does not have assets. So, what are you going to hypothecate other than your IP? And banks don’t accept it.

So, a mindset change, at least in the case of nationalised banks, has to be driven as a policy initiative. Or create a separate corpus where debt financing from nationalised banks becomes easier to access than being cornered by the online aggregators and the other dubious guys who presumably give you debt financing but typically at usurious rates.

We also need to educate most of the founders, especially in India and around the world. Money is not something that somebody turns on the tap and keeps flowing (laughs). At the end of the day, somebody is accountable for that money. I have had founders who have asked me, 'Ajith, can you believe it? They are asking me for a path to profitability!' My question was, 'You didn't have one?' 

People pick up some of these things from the United States and think that that works perfectly well in India. Business, by its very definition, has to generate a profit. That profit is what you use to plough back in to continue to grow.

And finally, you set up shop in Bengaluru, obviously because you were here from before, but what’s been your experience running the startup in Bengaluru?

I've had zero interference from any state government entities, positively or negatively (laughs). Let me put it that way. I have lived here (Bengaluru) since 1993. Till date, not one person has asked me for a political donation. Irrespective of parties. I've had the opportunity to meet with Siddaramaiah sir, Yediyurappa sir, and a couple of other bureaucrats. All of them have been nice, warm, always encouraging.

We are on the cusp of a giant leap. With a willing government, enabling ecosystem funding, a vast pool of talent, necessary natural resources, and a willingness to try, come what may, I feel that in my lifetime, we will become a developed country.