Full Transcript

Dhruv Sharma: This is TON117, we are streaming live and today Utsav and I are chatting with Roshan Raj, who's the founder of a company called Blurgs AI and they're working in the maritime domain. Roshan, welcome to the show, welcome to The Offline Network.

Roshan Raj (Blurgs AI): Thank you so much Dhruv and Ustav for having me, really pleasure to be here.

Dhruv Sharma: It's our pleasure. Tell us more about the company, Roshan.

Roshan Raj (Blurgs AI): Great. So, Dhruv, basically what we do is we're building the Google Maps for the maritime industry where we are bringing all the vessels, anything that floats on the ocean to a maritime map that you can track, you can trace, you can find out about anything and everything that is happening in the ocean world at any given point of time and we want to make it available to all the stakeholders in the industry so that decision making and all of those things can get better.

Dhruv Sharma: Excellent. So, you said smaller vessels, is there a reason you brought that up first?

Roshan Raj (Blurgs AI): Yes, because the larger vessels, the commercial vessels we have been catered for right now, there are about 1.5 to 2 lakhs larger vessels, but there are about 5 million plus smaller vessels, which currently does not exist on the maritime landscape or a digital landscape.

Dhruv Sharma: So, if you're the captain of a large Mersk vessel at sea, you have radars at your disposal and maybe the shipping line gives you a terminal on which you can trace all of the ships in your shipping lane, etc. But you're saying if you're the captain of a fishing trawler, you don't necessarily have that.

Roshan Raj (Blurgs AI): That's correct. There's basically something known as AIS. It's basically the GPS of the maritime world. Only the bigger vessels have it. The smaller ones don't have it.

Dhruv Sharma: What's the full form of AIS?

Roshan Raj (Blurgs AI): It's Automatic Identification System.

Utsav Somani: And what other imagery, sensors is required when you're building the solution for the remaining 5 million ships which are not on this sort of flight radar for ships? What other imagery, sensors, satellite data is required to build something like Tugs?

Roshan Raj (Blurgs AI): Correct. So, first of all, the bare minimum is to have a device that is compact, which can go on a smaller vessel. So, if you look into why it is not there in the smaller vessels right now, because the physical dimension of an AIS device is pretty big and it only justifies the space and the electrical requirements, all the network requirements. It justifies it to go on a bigger ship, but the cost and the size does not allow it to go for a smaller one. That's the first problem we're trying to solve is to make it smaller so that it can go on a smaller vessel. The second thing, obviously, there are more aspects to that. Now, think about like a dash cam for vessels, right? We have dash cam for every vehicle out there on the roads right now. And there are a lot of downstream benefits that you can get once you have dash cams, whether it be insurance game or any other those kind of aspects, right? Something similar is not available for the smaller vessels and we want to build it for them.

Utsav Somani: And so, I mean, you're miniaturizing this AIS as well. So, you're doing hardware also along with the software. Correct. And this software must be feeding, I mean, getting feeds from so many different sensors that all of them have different refresh rates. So, is that a secret sauce that you're building in blogs where your AI platform is combining all of these different signals with different refresh rates, different times and giving an output with some level of certainty on the location?

Roshan Raj (Blurgs AI): Yeah, that's broadly what we started the company with, right? If you look into the maritime landscape right now, it's a very traditional industry, meaning a lot of things have developed over a period of maybe 100 years, right? So, there are tech, which was, you know, some tech is 50 years old, some tech is like two years old, which is very fresh, like your satellite data, et cetera, which did not exist 20 years back, right? So, now, when as an operator, you're working in this industry, think about somebody, right? They have to know what was working 50 years back. They also have to understand what's the new tech and marrying them together is what the platform is doing so that a new guy who's operating in the systems, they don't have to look into independent systems. We provide them through, obviously, machine learning AI, what is coming up into this world, but providing them with an analytical tool that makes their life easy when they go out into operation every day.

Dhruv Sharma: So, Roshan, we used a couple of analogies to help understand what Blogs does, Google Maps and then Flightradar. So, when I use Google Maps, for instance, all it tells helps me with is navigation and tells me a little bit about the traffic situation where I'm headed. Flightradar, depending on the subscription you have, will give you a few extra parameters about the aircrafts that you're looking. What is it that mariners who are using the Blogs technology, what do they use it for? I mean, they have lots of navigation instruments. Do they use it for collision avoidance and things like that?

Roshan Raj (Blurgs AI): Right. If we think about both the analogies that I gave you, first thing first, Google Maps is able to give you accurate prediction because it has all the digital data of majority of the vehicles which are there on the road. So, that is the first problem we are trying to solve. Can you identify or get the remaining 5 million vessels who are just not in the digital landscape to the digital map first? Then it starts making sense to give a prediction. So, we are solving digitization first, then AI prediction and all of those things comes after that. So, what flight data type of analogy if you put into our platform, once they're digitized, now you can actually get to know about every vessel that is there. Currently, the visibility does not exist. There are a lot of national security related things that have happened in the past in India as well. It is primarily because we are not able to trace back the origin of any vessel and maritime is very complicated. You can just launch from any coastline. It is not like you need an airport to take off. You can literally leap into the sea from any point and go anywhere and come back to anywhere. So, that creates a pretty complicated landscape, maritime landscape, which just makes it very interesting to solve.

Dhruv Sharma: Can you also maybe for the benefit of our listeners, just tell us how oceans are navigated? Because if someone knows nothing about this, they will think that captains have free will to take the ship out on a whim. But really, there are these maritime highways if you see the traffic and density. Talk a little bit about that as well.

Roshan Raj (Blurgs AI): Right. So, if you look into the entire maritime landscape, you have to optimize on something. You can go anywhere. Does that mean you go anywhere? No, you don't. You basically optimize. What do you optimize for? You optimize for fuel or you optimize for time. There's no point of spending more time on the ocean. It's like a flight, right? What do you do if I spend eight hours instead of a six-hour flight unless there is obviously a deviation that I have to take for weather or any type of geopolitical reasons that I cannot take that route, right? So, they are optimizing for that right now. And that has created some trade routes. If you go into the internet and you search for that, then find out what are the given trade routes that any ship from point A to port A to port B, if they go, what do they take? But that is another traditional route that you have been taking. It does not account for real-time challenges. Like if a traffic builds up, there's no way you are getting updated that if something has come up, I should take a different route because they're following the traditional path, right? So, that is where real-time technologies like AI and real-time tracking and all of those things can enable better decision-making and faster optimization. Also, one very interesting thing that has come up recently is sustainability. Earlier, people used to optimize for time, like the faster I reach, but now they have to also optimize for fuel. So, going fastest does not mean you are optimum in fuel consumption. And there is a, when you drive a car, there is the optimum economy thing, right? So, you have to marry both. You reach early, then there is nothing you are doing at the port because the berth is not empty. Or you reach late, and then also nothing can be done. So, that marriage of when you go to a port, what is available, what is not available, that entire decision tree of how you should travel on the ocean, that is something that is very interesting and that is evolving with technology coming into picture right now.

Utsav Somani: So, what is proprietary to Blurgs right now? Because, I mean, say, how can a startup from India digitize some ship sitting in Norway or sitting in America? Like, how do you sell the technology to them?

Roshan Raj (Blurgs AI): Yeah. So, for us, the broad data line is the data models that we have built. Now, if you look into the weathers, the patterns, right? The maritime behavior itself, it's very complicated. Not something that you build in India, it's going to work in Norway. It's just not going to work because the wave patterns, the weather patterns, the routes, the island, everything is pretty complicated there. So, we have built a foundational layer on which we customize our model to every territory that the software is getting deployed. So, it's a foundational layer. Now, if you are deploying it for Norway, the model will get customized and trained on Norway-centric weather data, past pattern data, and all of those things, so that it starts to give accurate results in Norway because they have different tides. So, you cannot probably go at 20 knots, right? In India, it's easy to go at 30, 35 knots because the tides are not restricting your movement. It's pretty different. So, that is where the proprietariness comes for us. That's what we have been selling to the customers as well.

Dhruv Sharma: You were also telling us earlier about something about, you know, how in many places in the world, even rivers are navigable. And so, I think you have an application for inland water transport as well.

Roshan Raj (Blurgs AI): Correct. That's something you're working on. If you look into it, you probably would have seen in European markets, their inland waterways, especially in Germany and all, if you go, they're pretty, pretty commercialized. There are, you know, ferries and jetties that runs on the waters, right? On the rivers. But it's not pretty common in India. Primary reason is the tricky nature of our rivers were heavily dependent on monsoon, right? And running on ocean is fairly simple because it's open, but a river has its own course. You just can't go the way you want to. You have to follow the path that the river takes. And it's more complicated because, you know, there are seasons where the water levels are really good. There are seasons where water levels are too good. You can't navigate in there, right?

Dhruv Sharma: The sand is flooding right now, even though Brahmaputra is the widest waterway we have.

Roshan Raj (Blurgs AI): Correct. So, yes.

Utsav Somani: Sorry, please.

Roshan Raj (Blurgs AI): So, that makes it more complicated. So, it's a completely different, you just can't take up what we have already built and put it into rivers. It's just, I would thought it will work. It just did not work. So, we have to rework everything. And also pretty interesting data point, like national highways. India is also investing in national waterways. So, we have identified Ganga and Brahmaputra to be the top two right now, on which work has already started. The other additional rivers will also get added into this policy. So, it's a pretty interesting time to also look at how the waterways are getting evolved in the next 5 to 10 years.

Utsav Somani: I think, I mean, going a little bit off topic, but one fact that we were discussing before, I read the flight radar statistic that highest number of commercial flights day before yesterday, 150K flights, I think, which was the maximum ever recorded. What's the statistic like this for the maritime industry? And what's the state of the maritime industry versus commercial airlines?

Roshan Raj (Blurgs AI): Right, right. So, on a commercial airline, if you look at any given point of time, there are about 12 to 15K flights which are active. While for a shipping industry, it's about 1.5 lakhs, which are active on the digital landscape. But there are about 5 million vessels which are also not on the digital landscape right now. Think about there are another 90% flights which are moving on the skies and you cannot just see them on flight radar. That's a scary situation now, right? So, that is where the entire game completely changes for the maritime industry, which is just very traditional. They have to be, you know, there are smaller fishing boats, smaller vessels, who are just doing their day-to-day activities, their livelihoods and all that makes the entire game very interesting.

Utsav Somani: And congrats on the recent round that you've announced led by Pravega and Shastra as well. So, what's the business model for Blurgs? Like who are you selling to?

Roshan Raj (Blurgs AI): Yep, great. So, we have two different verticals for Blurgs. We have a defense arm where we're working with defense industry to provide them with intelligence to safeguard our national interests. Then we have a non-defense arm where we're working with ports, fisheries and shipping companies where we're providing them with operational efficiency data and information like that, right? That's the core focus. The funding was primarily raised to sort of go into international markets because we have formally established ourselves in India to an extent, but taking the same business model, going to Southeast Asian countries and Middle East right now. Also, there are some plans to go into European market as well in the next 18 to 24 months.

Dhruv Sharma: Roshan, because you'll spring up like defense applications, one of the questions I would have is, you know, a lot of seagoing navies around the world, including ours, have platforms like P8, like Poseidon from Boeing, etc., which have the most sophisticated onboard sensors and they can do a whole bunch of data fusion also while they, you know, surveil a certain area. So what gaps in capabilities are, like, you know, still remain that startups can actually go after, like realistically speaking?

Utsav Somani: I mean, I remember the name Poseidon because of the movie yesterday, Odyssey.

Dhruv Sharma: Yeah, Lord of the Seas.

Roshan Raj (Blurgs AI): Yeah, I think that's an interesting thing that even we identified, right? If you look into Poseidon as a it's one full stack system. Now, when you look into decision making by national security agencies, right? If Poseidon, let's say, gives you an analytics key, hey, we have identified something here that itself might be good enough, but not great enough to take an action. Usually how decision making happens is you need multiple sources of truth. Yes, I'm getting some information with Poseidon, but I need to get additional information from maybe data that I'm receiving from a nearby ship that I have, right? Or maybe from satellite imagery, or maybe from the radar data that one of my internal vessel is collecting, right? So now this becomes multiple layers of truth. And that is when a serious decision is made in the different forces, because they cannot go wrong. The question of having one 99.99% accuracy does not work in defense, you have to be 100% accurate, right? So that is where companies like us come in, we build multiple sources of truth, meaning Poseidon's data becomes one source of truth, then we work with other data modalities, other sources of data, and fuse them together to give it common decision hierarchy that hey, we have identified this information, but it is also validated from seven, eight or 10 different sources.

Dhruv Sharma: Because it is important to triangulate before you you know, move people into data. I think Poseidon can indicate that you give you an early warning signal that there's signs of trouble somewhere, even the first thing they will do is just dispatch a warship to go take a closer look. Right? Before doing anything. Yeah.

Roshan Raj (Blurgs AI): Yes. So that that contextual information of which warship to send, right? Who is closest, which what can be the my next action. So that kind of information also becomes very critical when you are short on time.

Utsav Somani: So I mean, of course, we can't talk about the defense deployments on air. But maybe tell us a little bit more about some of the private stuff that you've done, what workflows have improved for companies that you deployed this at?

Roshan Raj (Blurgs AI): Yep, great. So we have done a couple of things. First, I'll talk about one activity that we have done for shipyards. This is primarily one of the largest shipyards of the world. It's Dubai Maritime City. What we have done for them is now they have so many ships, etc, that is coming into the boats. Now they had a very strict way of how they wanted to operate. Now what happens because of that is, if you look into a shipyard, it has let's say, parallelly you can repair, let's say, 20 vessels, right? You have 20 boats. Now what was happening is, you know, they had no visibility about which boats are getting utilized and which are not, because of ease of operation, the closest ones to the to the coast or to the port was getting utilized more compared to the ones that were far away. Easy, just, you know, convenient. What happens because of that, those are getting sort of used, overused, so the materials, the equipments used on those boats are getting spoiled faster compared to the other boats. So, but their maintenance activity is flat, that in six months we'll do a flat maintenance. What happens because of that, some does not need maintenance and some required maintenance probably two months before. Now there is a breakdown, and if there is a breakdown, there is a huge revenue loss and the operations and all of those things commonly stopped. The second thing we are also doing is for the Indonesian government, in which we're helping them identify or helping their fisheries become more, or fishing activity become more, you know, centralized, meaning when a vessel goes into the ocean, everything that they're catching, it's digitized, it's data logged, and the information is captured in a traceable format. Meaning, let's say somebody in US or European market is consuming a fish, they exactly know that this fish was sourced from this island, off the coast of this, this date, this time on this, and this is also approved by Indonesian government that it was sustainably sourced. Meaning the entire traceability, if I tell you, I'm giving an Kashmiri apple, how do you know it's Kashmiri? Or it's not sourced from some other place, right? So that traceability of where this value is coming from or where this product is coming from, that's something we're doing for the fisheries of Indonesian government. Again, multiple sources of truth to enable this concept.

Dhruv Sharma: Great. Roshan, maybe the final question I have for you is, what's really happening in shipping in India? Like we've privatized a lot of ports, we're doing a lot of ship building, we also have ship breaking yards, give us a sense of what's really going on.

Roshan Raj (Blurgs AI): Yeah, it's pretty interesting time in India, honestly, in the maritime space. So we are currently being positioned as to become the one of the world leaders in maritime ship building. India is one of the largest population as seafarers, meaning we have a lot of population of Indian population that is going into the shipping lanes and working for the shipping industry, right? But we're not in the top ship building countries of the world. So Indian government is investing massively like about, I think the recent update came out 70,000 crore plus 70,000 lakh, sorry, 70,000 crore lakh, something like that. I'm missing the context, but a huge amount of money is being pushed into maritime industry for building the entire supply chain for Indian ship building. So that would become one of the top ship building companies globally. So that makes up an interesting landscape of number of vessels that is going to come out of Indian ocean, Indian coastline. The second part is also government is heavily investing in transformation of inland waterways. That's the second part. The third is there were five builds, fundamental builds that have been revamped, which were centuries old, right? These were like early 1900 builds and all of those things, which was never changed, right? They have revamped them to categorize it to a way where it can be consumed or it makes sense in today's world, meaning massive boost to coastal shipping economy and all of those things are becoming pretty interesting in India right now, at least what we have seen. And this has trickled down effect not only in India, it's the entire Indian ocean region, right? If you look into Middle East, interesting things are happening there. If you look into Southeast Asian countries, pretty interesting things are happening there, right? We are building new ports, there's a Wadwan port coming off. We are also revamping the entire Andaman Nicobar Islands and that cluster so that we can control the trade routes and all. Very interesting time of what is possible in Indian maritime future in the next 10 to 20 years.

Utsav Somani: All right. Thank you so much for coming on our show.

Roshan Raj (Blurgs AI): Thank you, Shyav. Thank you, Dhruv. Really pleasure having you, being a guest and sharing my story. Thank you so much. Good luck.

Utsav Somani: Super exciting. Thank you so much. All the best. All right, listeners, moving on to our next guest. We've got Ajaz from Novite Materials. I hope I got the name right.

Ajaz Khan (Novyte Materials): Absolutely, yes. You're one of the few people who did get it right the first time. Amazing. What does it mean? So it was basically a wordplay on novel new data. So Novite, where white being, white e being bite. So Novite, Novite. So that's about it.

Utsav Somani: Let's introduce the company also, now that we've given context about the name.

Ajaz Khan (Novyte Materials): So we are an EF materials discovery company with a focus on manufacturability and a scale up of new materials. Typically, whenever anybody tries to discover a new material, it usually ends up just trying to make it in the lab or even computationally figuring out whether it's going to be stable or not. But at the end of the day, when you try to create a new material, how are you going to earn money from it unless you actually do it at scale? And that's where Novite actually lies, trying to take you from first thought to your first product end to end.

Utsav Somani: So when a customer comes to you, what exactly are they asking for? Do they come with a vision in mind about a product that they're building or do they come to you that, hey, this is a material science problem that we want to solve? Can Novite help us solve that?

Ajaz Khan (Novyte Materials): So the vertical that we're actually primarily involved in is the speciality chemical space. And there are multiple issues that they need to solve for considering the number of dynamics that are playing out here. Number one being whether should we optimize our existing products? Should we premiumize it for a larger scale? Or does the process chemistry needs play because of supply chain constraints? Suddenly a country X one day says we're not going to supply a material to you anymore, which was very much part of their formulation. A company stuck. Or trying to figure out how exactly a new material can be put into their process to make it cheaper or optimize it further. So they come to us with one of these four issues most of the time. And typically, how their workflows would go is one to two years in trying to figure out whether it's possible or no. And optimizing that is another two to three years. And then taking that at scale is another ballgame where we see multiple 95% of the materials fail at that point.

Utsav Somani: So, or give us an example, like, I think that'll help us understand better, like any customer that you can mention, or maybe just sanitize the name of the customer and just give us an supposed hypothetical example.

Ajaz Khan (Novyte Materials): So a specialty chemicals manufacturer wanted to replicate the product or the material made by a global player, someone who's Forbes 500 and insane person company, they were stuck on the replication of that particular material for four and a half years. And they were not able to do not even get it to lab, because it was failing nonstop. The they just had to upload something called as a technical data sheet, which is what tells us the constraints or the properties of a particular material that needs to be hit to actually make it. And we were able to design, synthesize, optimize and take it to scale within three and a half weeks, compressing that entire timeline. So and whenever we usually meet with clients, there's a lot of skepticism, it's not gonna work, it's, you know, it's not possible at all, because the chemical industry is, you know, for the lack of a better word, a caduce industry. It's filled by old people who, you know, are these manufacturers, and they don't believe AI can work. So whenever we are working with them, we usually play the devil's advocate, you tell us a material or a formulation that you're stuck on, right? And you see how much time it took you. And then we do the same process for you within the demo itself. And that's what, you know, allows them to convert significantly faster.

Dhruv Sharma: And just, you know, a couple of episodes ago, we had somebody from the drug discovery world. And the moment you said materials discovery, suddenly I started drawing like analogs between the two. So what is it about AI that makes it, you know, just a great technology to use to generate candidates, whether it's in drugs or in materials? I mean, in their case, you know, they'll isolate protein sequences and go from there. With your example of specialty chemicals, talk to us a little bit about how you generate a candidate. And then once you get it out of the lab, how do you actually bring it into production and manufacturing?

Ajaz Khan (Novyte Materials): Okay, so for example, when I was doing my master's, right, I came to realize that to understand how a material can actually be formulated, or how a material can be created, the current process is actually just make it in your lab. And then you see whether it is stable or not. And post which you have to see whether it has the properties for your particular process. So what I did is instead of that, why don't we try encoding individual elemental properties into a node? So why does sodium react with chlorine? Using as they exist in nature? As they exist in nature. So like, so why does sodium react with chlorine to form salt, right? There's a positive charge of sodium is a negative charge of chlorine. When you combine the both of them, you have a total charge of zero, which means that the material is stable. And you can map out 100 such factors, which correlate to the stability of a particular material class based on just these. But the problem is when you're using large language models, they hallucinate. Because if today I'm an apple, and tomorrow I'm an orange, and LLM is going to remember both. But in science, you can't do that, because that will lead to hallucinations and that will cause problems down the line. So what I did is I made sure I created a neural network at its core during my academia days, which map out individual properties, and then I let it loose on the periodic table. So instead of training it on a data set, instead of training it on a particular sequence, whatever, because if a data set has ABCDE, how am I going to find a material at XYZ? So I let it loose on the periodic table, I let it figure out what is going to actually happen, add it in a reinforced learning loop for the system to fundamentally understand what is leading to the stability of this material in the first place.

Utsav Somani: But added in some external element, I mean, external data sources that you need, right? I mean, real world examples of what's out there already and what is already built and solved for.

Ajaz Khan (Novyte Materials): What we really need is more the geometry of the material rather than the chemistry of the formulation. For example, when you're dealing with, say, something called as a perovskite material, P-R-O-V, all right? These are materials that are not dependent on the element, but they're dependent on the crystal class. So for example, cesium, lead, and bromine, right? Now this is a chemistry. But if you strip out these elements, you can write it as ABC. That is what we constitute as the geometry of a material. So the only thing that we need externally is the geometry of the material that we're dealing with, which would either be the constraint on the application given, or we just let the system understand what is inherently becoming a stable material based on its combination. And post which, we added a filter of something called as density functional theory, which is quantum chemistry. And that looped in. So usually when you discover a single new material, you end up getting a PhD. The system that I orchestrated discovered 39 out of which 24 ended up being stable, keeping us at a hit rate of 68% over the industry hit rate of 0.5 to 2%. And yeah, that's how we started off.

Dhruv Sharma: And you know, I was just saying Ajaz, well, that was illuminating. I can assure you we have very, very few scientists in our listener base. I think one of the things we'll get you to do maybe is outside of specialty chemicals, let's just talk to us about advanced materials, specialty materials in our everyday lives, right? Things that we've been using forever, fiber, ceramics, super alloys, etc, etc, just start jogging our minds so that we can follow and track what you're seeing a little bit better.

Ajaz Khan (Novyte Materials): So the most interesting speciality chemical that I am pretty fond of is something called as colloidal silica. So you see, every single structure that you see, I thought you were going to say very quick for some reason. You can draw a parallel to that. It's sort of the same. But colloidal silica is basically a chemical that is used in the cement and construction industry. And typically, when you are constructing a building, you have to put something called as a beam, which is the slab that goes on top of it, having a load bearing capacity of say, 30 to 35 megapascals. When you construct high-rise buildings, you have multiple factors at play. Number one, the load increases. Number two, wind is hitting from one side, so sway increases. So rather, and builders are not going to increase the size of the beam because it's going to cost them valuable real estate. So the specialty chemicals that actually go into this increases the load bearing capacity from 35 megapascals to 120 megapascals. And how it does this is it tries to account get into the science of this. All right.

Dhruv Sharma: This is great, right? Because even for instance, we didn't at one point, we didn't have a lot of steel in India, which is why we went the RCC way, right? So you innovate. If you don't have something that's found in nature, you innovate on what you can. By the way, speaking of swing, do you know some builders actually put pendulums at the top of their skyscrapers just to keep them stable? This was news to me when I read it last.

Ajaz Khan (Novyte Materials): So it's because, yeah, sorry.

Utsav Somani: Yeah, what was the reason actually, from my understanding?

Ajaz Khan (Novyte Materials): So when you're trying, when the sway is particularly happening in a building, right? They try to ensure that the balances are loaded in from different sites. So load bearing on the side goes in from that side. That's about it.

Utsav Somani: Wow. And I mean, so who keeps the IP in your case? Like, I mean, who actually owns the output that is coming out and getting used in the final products?

Ajaz Khan (Novyte Materials): So we split the IP ownership in two cases. The first case, when the platform is deployed locally within the company, they own the IP for any optimization, any product development or, you know, formulation that they do based on their existing KPIs. If they want to discover a new material end to end, which is where large scale simulations go into order, they're trying to discover it and Novite synthesizes it in its own lab. That is the only time Novite owns the IP end to end, where then we license it to the company and they pay us a royalty on the entire deal done. Otherwise, the IP is owned by the company.

Utsav Somani: And eventually at scale, like maybe 5, 7, 10 years down the line, if the hyperscalers are unable to catch up to this, or the models won't improve that much, or you don't have any other competitors globally, you can own like a port and many others can exist, but you will own a portfolio of like solid speciality chemicals.

Ajaz Khan (Novyte Materials): Absolutely. And that's the entire goal of it. At the end of the day, the final goal of Novite is to become a foundry. The only reason why multiple chemical industries, so DuPont became DuPont because of one chemical. All right. And if Novite is able to own multiple such large scale IPs, it would make significantly more sense for us to own those IPs, create spinoffs of our own and then verticalize one by one. But to get to that point, you need to fundamentally de-risk the whole stack to ensure that at day one, you know whether it's going to be manufacturable or not. And that starts with us working with manufacturers right now.

Utsav Somani: Exactly. So that's interesting. But do you file patents for this? Like those stuff that you own?

Ajaz Khan (Novyte Materials): Yes. Yeah, yeah, yeah, yeah, yeah. We do, we do. In fact, we recently signed a manufacturing and royalty agreement with a company called Kemper Speciality Chemicals. They have a manufacturing arm in the US called Michigan Additives. The IP is Novite and the job descriptions are very clear. Novite discovers and synthesizes, they manufacture and distribute.

Utsav Somani: And do you have plans of going into manufacturing?

Ajaz Khan (Novyte Materials): Not yet. Not yet.

Dhruv Sharma: Which actually brings me to a question, which is along the way, if you actually stumble upon a breakthrough discovery, what are you guys going to do then? We'll then get into manufacturing.

Ajaz Khan (Novyte Materials): Because see, manufacturing is not easy. Manufacturing is a game that causes, has a lot of variables at play. So for example, when you're in a lab and you're trying to figure out what's going on in a beaker, you're at 100 RPM, you have a 50 ml, you know, beaker. But at scale, you're doing it at say 1000 liters. And then the speed of your agitator is now 8 RPM. So the AI needs to fundamentally know how to map that out at that scale. And it gets harder and harder, the harder the process chemistry actually lies. So yes, so at that point of time, if you're able to discover a breakthrough material, yes, we will verticalize, but we need it to be de-risked enough to actually get to that scale. The whole point is to do this for multiple new chemicals, step by step.

Utsav Somani: What industries or products are you most excited by, where you think your product can play a part in discovering new speciality chemicals as inputs?

Ajaz Khan (Novyte Materials): So within the speciality chemicals as well, I would say catalyst is a major space that we're actually very much interested in. Because it's the one material class that is, that can have multiple applications across various processes, and make them cheaper, while adding nothing much to the cost. Right? Because it's another concern. So that's one thing that we're really interested of. In fact, one of the first few things that I actually was running a simulation on was something called as a covalent organic framework, which I felt would lead up into building a catalyst for a particular industry.

Dhruv Sharma: Globally, who's doing this Ajaz?

Ajaz Khan (Novyte Materials): So it's very sad to say, but in India, so we are the only people working on AI for materials at scale. Globally, we have people like periodic labs, orbital materials. So AI for materials is really popping off in the West with billions of dollars in valuation. The one major differentiator that I would see between them and us is that no one is again, actually focusing on manufacturability of the materials. Every single one of them is going into creating the next God material, or trying to figure out that one material that would, you know, change everything on earth. But when you verticalize from day one, the entire philosophy is suppose the company is making a material that acts as a coolant in data centers. And tomorrow, and equally, if not more well funded industries taking all data centers to space, right? All your money's gone. So that's what people are doing globally. And yeah, and they again, working more on synthesis than manufacturability.

Dhruv Sharma: And talk a little more about the team as well, right? Like, who do you have working with you? And do you guys like, I mean, it's still very, very early, but right now, I'm assuming someone walks through the door with a spec and then you can, you know, you go from there. But at some point, are you guys going to like proactively start coming up with formulations, just be researching geometries, etc?

Ajaz Khan (Novyte Materials): 100%. So, so currently, we are in the middle of our second fundraise. And the whole point is for setting up our own lab outside. And so that's what we're going to do. And that is to work on this critical chemistries that you're talking about, set up the flywheel, ensure that the model gets better and better and smarter and smarter over time. Because what we actually also need is negative data. And negative data is experiments that are experiments that have failed. So, because if a material is positive, and it's working, it doesn't really teach the model anything. But if it knows why it failed, it is going to get to your successes significantly faster. And that's that.

Utsav Somani: And if India wants to fulfill his dream of becoming sovereign in, I mean, material science explorations and stuff, what is the current bottleneck in our ecosystem? Like is it labs? Is it manufacturing setup? Is it something else?

Ajaz Khan (Novyte Materials): It's the incentive of chemical industries to it's the fail rate. That's what the bottleneck is right now. Because the Indian chemical industry right now has very little patience. Why does materials discovery take a lot of hands? Because it costs a lot of money. It costs a lot of money because it takes a lot of time. It takes a lot of time because there are a lot of failures. And currently, we have people like CROs, whose entire job is to hire like 70, 80 people to endlessly perform experiments, get to a particular molecule and try to sell it, whether it be globally or whether it's in India. AI is able to fundamentally offset this entire cycle. What would fundamentally take a person a few years to even understand what the problem statement would be? For example, if somebody wants to create a paint that has a replacement for...

Utsav Somani: I mean, I'm remembering what Dhruv, the analogy with drug discovery as well. That's why we're selling generics globally to the US and stuff and exporting that mostly. But if we use AI and startups like yours succeed in different vertical applications of discovery, I think it can be fascinating for India as an export manufacturing capability development.

Ajaz Khan (Novyte Materials): Not only that, but look at it from this perspective. If we are able to utilize the resources present within the nation itself, rather than be dependent on external imports. For example, there's a chemical called antimony trioxide. Antimony trioxide, so you see every cable that's in and around you, it's used as a fire retardant in there. And suddenly one day, country X said, I'm not going to give it to you anymore. The cost went from $6 to 70 a kg. So now India came under significant constraints. Now, if we had a formulation that was dependent on a domestic supply chain, India could very well become a sovereign manufacturing ecosystem, fueled by AI to supply globally. And that's where the foundry, that's where new verticals. This is one industry that's untouched by even digitization. People are still doing it on pen and paper. A senior scientist leaves a company, all the data goes with them, because it's present in their mind. It's not even written anywhere. So there's a really big opportunity for India to become a leader in this space with how things are. And that's really what we're aiming for.

Dhruv Sharma: I have one slightly twisted question, Ajaz, which is in this category, how hard is it to reverse engineer something? I mean, we're talking like at the microscopic level, how hard is it to see and work backwards from there?

Ajaz Khan (Novyte Materials): So initially, it was really hard, because most of these things are usually barricaded by not only the material, but also the process chemistry that goes behind it. So you can make an app, you can make an app, for example, like you have an app, you can see the output, but you don't know what led up to it. That's also really well hidden behind, you know, within a company as well.

Utsav Somani: How we sort of try to compete with that is try to draw a Coca-Cola recipe like out in the open, but nobody can reverse engineer. I think they've not been able to do it. Yeah.

Ajaz Khan (Novyte Materials): So, for example, what we did is, because we already had, based on what I did during my academic years, we could always draw parallels. So there is a chemical called HNO3, which is nitric acid. It's very well documented. We have an excellent way of manufacturing it the way it is. But its sister family is not so much. So how do you manage to, a human being would have to spend years actually trying to figure out what parallels there are. But with AI, it could even easily draw what parallels there are, what contributions there are, what it could take from that process and try to get a new process for you. And for an industry, it's anyways going to be an iterative process where they are going to have to do it a thousand times with NOVA. Now they're doing it, say, 20, 30 times. And now they're able to fundamentally reverse engineer whatever they'd like.

Dhruv Sharma: Makes sense. I'm going to ask a quick follow-up, which is Ajaz, let's assume your best friend also wants to start a materials lab and he promises you he's not going to touch specialty chemicals, but asks you, hey, what other opportunities can I go after? What are you going to recommend to him?

Ajaz Khan (Novyte Materials): Within the materials, I would say to go into semiconductor chemicals.

Dhruv Sharma: So that comes under energy storage? No.

Ajaz Khan (Novyte Materials): No. One thing that people don't really, one industry that I'm really interested in is polishing. So when you are trying to create these really complicated pieces of equipment, which is whether it be the rotors of a plane or whether it be an electron microscope or whether it be a semiconductor wafer, you need to have a very fine level of smoothness to the surface. It's a very fine level of precision. How do we get that? We can't just take a sandpaper and drill it, right? No, it's not going to work out. So at that point of time, there are a set of chemicals that you have to actually dip all this and then vibrate it to create a smooth surface as a whole, which looks flat even under an electron microscope. And that is what people are unaware of because that's the industry that is used everywhere from semiconductors to electron microscopes, to aeroplane blades, to even reactor systems. I'd suggest him to go there.

Utsav Somani: All right. Thank you so much for coming on our show. If you find the super molecule, I think give it to me. We'll set up a plant to manufacture it. Wishing you all the best. Thank you so much. Thank you. Bye bye. Have a great day. All right, listeners, that's it from us. Wednesday is stream 118. We'll see you on.