Full Transcript

Dhruv Sharma: Hey there listeners. Happy Monday. This is TON 115. We're streaming live. And just this weekend, we don't get to say this very often, but a startup truly created history. And as you might know, the startup we're talking about is Skyroot Aerospace, which after it launched Vikram 1, their first rocket into low earth orbit, India became one of the only three countries in the world that have domestic private orbital launch capabilities. So obviously, there was a lot of excitement around this. And, you know, congratulations poured in from every single corner. I think the ecosystem was super, super excited about this. There is something about space, which, you know, makes any accomplishment, not just about one company, but really the whole country gets behind them. And, you know, the other thing that was great about this launch is it's very rare in space for any company to get everything right in the first go. And that's exactly what Skyroot managed to do. This was a flawless maiden voyage. And the rocket took along with it a very diverse manifest of payloads. There were at least six items that we're aware of that included a satellite they've made by themselves. They carried a satellite for another Indian domestic. It wasn't a satellite. Maybe it was a satellite pathfinder, but for a domestic space company, even for an international customer, they carried a robotic arm in space from someone who's actually been a guest in one of our previous episodes. That's Cosmoserve, a robotic arm that will just find a way to catch space debris. And then they carried some keepsakes of deep sentimental value for all of us. That included a handwritten postcard from the Prime Minister. It included some items. I'm not exactly sure what they were, but, you know, miniatures of C. V. Raman, Vikram Sarabhai, and Dr. A. P. J. Abdul Kalam, all of them luminaries who've inspired generations after generations. And they carried, along with all of this, they carried a diamond with them, which was called Cosmos Bloom. And the company that manufactured that diamond, and their founders, in fact, are first guests on the show today. Before I introduce her, because we know you love space so much, and we love space a fair amount, we're actually going to do something special on space later this week. We'll put more information about that as soon as it comes together. But without keeping waiting anymore, let's welcome Sanjana to the show. Sanjana, welcome to The Offline Network.

Sanjana Tripuramallu (Cosmos Diamonds): Yes, it's great to be here. Hi Dhruv.

Dhruv Sharma: It's great to see you. And, you know, just as I said, just before we went live, you have arguably one of the most beautiful looking backdrops that we've, any of our guests ever. But tell us more about the diamond, Sanjana, how's it doing? What's, what news do you have about the diamond?

Sanjana Tripuramallu (Cosmos Diamonds): It's just absolutely surreal. So it's Cosmos Bloom. And can you believe it, like, the name of the brand is Cosmos. And to see something that we've made in the actual backdrop of Cosmos is a feeling I think I'm never going to get over. And it's a great story that I'll probably tell my grandkids one day. So yeah, the diamond is in orbit, and it's doing great.

Dhruv Sharma: That's insane. Tell us more about the story. How did this come like, how did this come together?

Sanjana Tripuramallu (Cosmos Diamonds): So it's, you know, it's, it's really a very simple, even one might even say silly story. So I've met Pawan in a business club. And so he was just telling, I mean, basic introductions, even in the beginning, it was like, Oh, why? I said, I'm from Cosmos. And he said, okay, this is a space company that I don't know. And they said, no, no, this is a diamond jewelry brand. And then he told me about Vikram. And I knew that they were going to launch a private rocket. So it was almost a silly idea, because you remember Twinkle, Twinkle, right? I mean, who doesn't? So it ends with up above the world so high, like a diamond in the sky. So their motto is opening up space for all. So I was asking, what are you going to carry? What is space for all? And then I was asking him, do you think you can carry a diamond? And he said, no, that's a very interesting thought. Why don't we do it? Because the first, so space has, so of course, along with all the satellites, you could always carry something of deep value and meaning, especially since it's a historic first launch. So he's asked, like, we want to do that. Why don't you take it up? And we thought, is this really happening? Like, is this really true? So we took it up. And then many months later, I can't believe that it's actually in orbit.

Dhruv Sharma: That is so, that is so awesome. Can you tell us, and you're a designer, right?

Sanjana Tripuramallu (Cosmos Diamonds): Yes, yes.

Dhruv Sharma: Can you, can you tell us more about the design of the Bloom?

Sanjana Tripuramallu (Cosmos Diamonds): Yes. So once we've come into agreement that we're actually doing it, there were many technical specifications that it has to adhere to. At the same time, because ultimately, it is carrying a meaning, right? It's a symbol, like literally anything you put on that rocket is definitely the whole world is watching it. And especially it's a message from India, because it's the first private rocket. And even the rocket is entirely designed in India. So, so we've all thought about this really deeply on what do we really do, because everybody's going to watch. And so generally, when you do jewelry, right, there is only one point of view. Let's say you're doing a pendant, you're only going to look at it from, from, you know, right in front of it. But then when you put it on a rocket, we don't know how is it going to be mounted? Where is the camera? How should it look? What if you're going to look at it from the top or from the side or from the three-fourth? We don't know, like, how do we design it? So there were many questions, many ideas, we've considered a lot of motives and a lot. So my main agenda was that whatever we put there, the whole country is going to watch it shouldn't be something very complicated. Anybody in the country, when they look at it, they should know what it is, which means it should be very simple. And it should be recognizable.

Dhruv Sharma: Yeah, very recognizable.

Sanjana Tripuramallu (Cosmos Diamonds): And yet it should have a significance. So since it's all it's carrying that Indian pride and historic mission, so I started looking into what can we put there, which will signify India. So Lotus was one of the option that had come because if you see our temples and architect, archaeologically, even if you go to a 7000, 5000 years old temple, you will see carvings of Lotus there. And usually Lotus is always associated with the God. Like if you see Buddha, he's carrying a Lotus. If you see Brahma, who's the creator, he's always on the Lotus. So there is that very deep association. And it's also the national flower. So we felt that and it worked in a sense that if you look at the blooming Lotus, it looks beautiful from all angles.

Dhruv Sharma: It does. Yeah.

Sanjana Tripuramallu (Cosmos Diamonds): Right. So we felt that that is the ideal motive that that can carry this mission. And that's how we came about the design. And then we built it layer by layer with lots of experiments and what settings would work because there's going to be a lot of pressure.

Dhruv Sharma: How many petals did the Lotus have?

Sanjana Tripuramallu (Cosmos Diamonds): So it has 32 petals and that's why 32 diamonds. And it was built in a way that it has a 3D structure. So that's why if you see the footage also, there is a view which is the most beautiful. The camera shows the Lotus being hosted from the side and then there's a sun that is just started rising and then the sun rays fell on it and it was just beautiful. So yeah, 32 diamonds overall about 17 carats. Every diamond stone size is different and it is in layers.

Dhruv Sharma: Wonderful. And I think this also sets pretty good ground for us to just talk about the space overall, Sanjana. Before you, we've only had one other founder building in the space and that was a really long time ago. I'm not sure if our current listener base has even heard that episode. So I think just like just tell us the basics. I think very, very often people just focus in when they hear a lab grown diamond, they focus on lab too much and not on the grown. So talk to us about the whole process. How long does it take to grow the diamond?

Sanjana Tripuramallu (Cosmos Diamonds): So there are many methods. So actually the industry has evolved very beautifully in the last 4-5 years. When I started this in 2021, back then lab grown wasn't really popular. It was only into exports. So if you are aware in general, since you're speaking of diamond industry, like Surat is a major hub, like 9% of the world's diamonds, not just India's, world's diamonds are cut and polished in Surat. So when the technology started improving, Surat has also taken on the responsibility to build growing hub. So there were many companies that started around 2020-2018. But 2021, there wasn't demand much, especially in India. And the also wasn't entirely stable. So around 2023 is when there was a bit of balance and equilibrium in terms of supply and demand and technology is rather advanced. So if you see today, I think there are so many startups that have come up in this space. And even from a market size point of view, today, natural diamond standard around $6 billion in India, whereas lab grown diamonds is around 400 million. It's still quite small compared to the overall natural diamond as a segment. But if you see the trend, right, look at exports. I think 2026 is the first year where lab grown diamond exports have exceeded natural diamond exports.

Dhruv Sharma: Is that a fact? Wow.

Sanjana Tripuramallu (Cosmos Diamonds): Yes. So if you see a couple of years ago in 2020 or so, lab grown is at 10% and 90% is natural. But this year, it has actually...

Dhruv Sharma: I think it will be fascinating to know how a city has reinvented itself. So talk to us a little more about Surat. I've actually not visited to date. I'm very curious. I'd love to learn how, you know, how they're adapting to this change and reinventing themselves.

Sanjana Tripuramallu (Cosmos Diamonds): So they have not adapted, they have literally started the change. And you know, Surat has, I mean, changed me in many ways, at least when I live in Bangalore. So the idea of a business or an idea of startup or how business works, I have met people in Surat who would probably, who's only studied till fourth standard and probably is a 60 year old and you would not expect and they would be running hundreds of floors of business. And they might not, they would be working with Russia and US and they'd be exporting to all these countries. And if you talk to them, they have understood this technology and started this hub way before anybody caught anything. There was no demand, nothing. But they knew that this is going to pick up and they're, I mean, yeah, amazing people, amazing industry. I often like to say, like, it's one of the only cities in India, which is measured by its GDP, like other cities being, you know, having a GDP count of their own.

Dhruv Sharma: And two other questions for you, Sanjana. One, is the technology still very expensive? Or is it getting cheaper as time is passing by? The second one is around market acceptance, how your consumers taking to it. One related question to market acceptance, but maybe after you've responded to both of these.

Sanjana Tripuramallu (Cosmos Diamonds): Yeah. So in terms of technology, right, I would say, like I was telling you, in 2020, the technology wasn't as accessible, it is still evolving, like the price, the cost of manufacturing has actually declined. And now it has reached a point where it is significantly improved already. It won't get any cheaper, it will not change much, because already a lot has happened. And now we are able to produce diamonds of really good quality in a considerably effective amount of time. So yeah, this change, so agro diamonds themselves are not very old. They've existed for a long time, but they weren't of the quality where you could use it for jewelry, because jewelry again, has really different standards. Diamonds also have other uses, right? So it's around 2020, that technology has become better. Again, if you have to go in depth, there are two methods, the CVD and HPHD. So yeah, there has been significant progress. And now you could say that it has already reached a point where it cannot get better. Or even if there is an advancement, it has to be significantly better to make a difference in the industry. So it's kind of reached at a point where it is great.

Dhruv Sharma: The other question I had with respect to like market acceptances, when a buyer, you know, walks in through walks into your store, does the sales associate talk to them the same way they would if they were selling a, you know, mine diamond? Or do they talk about different things?

Sanjana Tripuramallu (Cosmos Diamonds): Right. So coming to buyer acceptance, we need to understand this that India has always been a gold buying country. Because gold is associated with stability, we all know the price, they've a lot of, I mean, majority, right? Whereas diamond has always been an aspirational product, a young product. People wear diamonds when you want to tell the world that you have arrived. It has nothing to do with the cost, the security, the investment, none of it. Diamond is a purely aspirational product and it remains so. But then for a very long time, if you look at diamond buying in general, forget lab-grown, just diamond buying in general, it is very difficult to understand how this is priced or why is it that way? It's just, it's just, it's always been like that. And it always belonged to a class where maybe the top 0.01% would probably enjoy it. Whereas everyone else are just aspiring for it, but you can't get there. But then when lab-grown diamonds happened, it was like a very welcoming change now, because now you can actually own it. If you do a price comparison between the two and a quality comparison as well, since we're talking of both, lab-grown diamonds are at least 60% cost-effective compared to natural while actually being better in quality. That never happens, like at any given point of time, when the price is lower, it's usually because there is a problem with the quality or there is some difference, right? It's never of superior quality, but actually when you see lab-grown, because it grows in more controlled environment, the quality is better with lesser price. So yeah, it has been insane. Like today, the market acceptance is much higher, but definitely it came with a lot of education. Like for the question that you asked, how does the sales person respond? So as part of sales training to anybody who enters into the space, we have a dedicated area in every store where we explain the process and what are the differences and everything so that they feel safe before the purchase. But today, actually, a lot of them, we don't have to explain anything. They are the ones who will tell us, is this that? I've read about it. Today's consumer is very different.

Dhruv Sharma: I think we seem to be losing Sanjana a little bit over there.

Sanjana Tripuramallu (Cosmos Diamonds): Yeah. Am I audible now?

Dhruv Sharma: Yes, you are again. Sanjana, I got what you were saying as did our listeners. Maybe one final closing question for you. Now, the original Cosmos Bloom is in orbit. It's maybe never coming back home. Do you have a replica? Is there a second one that you also made?

Sanjana Tripuramallu (Cosmos Diamonds): Yes, I did. So actually, because I was supposed to go through extensive tests, like vibration tests and stress tests and everything, we made another one just to be sure that if something happens to it, we can immediately make another one because making it is a very long and tedious process. But as it would have it, the first one actually passed all the tests and is in orbit. The second one is currently in a safe. After this historic launch and the insane response and love that is coming from everybody, we're going to build it a small altar at our store and we're going to keep it forever.

Dhruv Sharma: Yeah. I mean, I know this segment was supposed to be focused more on jewelry. We ended up talking about space a lot. Maybe the second one will find great application because I believe there's another rocket launch coming up and there we're going to be testing reusability. So maybe the second time you send a diamond up, it can actually come back because that's the other advance that's coming next in space tech. But Sanjana, thank you so much for your time. Thank you for coming and sharing the story of the Cosmos Bloom. We'll make sure that more and more people get to know about this and get it, take inspiration from it. All the best to you.

Sanjana Tripuramallu (Cosmos Diamonds): Thank you. Thank you, Dhruv.

Dhruv Sharma: Yeah. Great talking to you.

Sanjana Tripuramallu (Cosmos Diamonds): Pleasure chatting with you.

Dhruv Sharma: All right, listeners, that was Sanjana. And now we're going to welcome Aniruddha of SwitchOn. Hi, Aniruddha.

Aniruddha Banerjee (SwitchOn): Hey, Dhruv. Hi. Good afternoon.

Dhruv Sharma: Good afternoon. It's great to see you.

Aniruddha Banerjee (SwitchOn): I'm great. How are you?

Dhruv Sharma: I'm great. Thank you so much for asking. Are you somewhere at a manufacturing facility right now?

Aniruddha Banerjee (SwitchOn): Not really. I'm back in office, but I was in the first half.

Dhruv Sharma: All right. Fantastic. So tell us more about SwitchOn.

Aniruddha Banerjee (SwitchOn): So we at SwitchOn enable large manufacturers to reach zero defect and zero customer complaints. So we are building an AI-powered inspection system that gets deployed in manufacturing lines of large, you know, CPG automotive electronics companies across the globe, you know, and enable them to reach zero defects and zero customer complaints.

Dhruv Sharma: Well, that's a very succinct explanation, Aniruddha. But who knows, many of our listeners might not come from a manufacturing context. So talk to us about this philosophy of zero defect. Like, why is it so important manufacturers?

Aniruddha Banerjee (SwitchOn): No, for sure. So, you know, very unlike how software behaves, right? Hardware products behave very, very differently, right? So, for example, typically in software, you know, you would be, you know, making a version, patching it throughout its lifetime. And, you know, that's kind of how it works. In hardware, unfortunately, the way patching works is that you have to get the product back into the, you know, manufacturing facility or the brand, and then reissue the product back to the customer, right? Or, you know, you essentially have complaints, people suing you, and it's not nice, right? So, manufacturing a product without defects is of utmost importance. In fact, it's so important that almost 25% of the typical workforce in the plant only works on quality and quality-related activities, right? And still, manufacturers lose in the order of about $500 million per year in what is called cost of quality or cost of poor quality, right? This is why it's very, very important for, you know, for AI to come and change the way that this works, because the way that it works right now is it's completely manual, right? So, imagine, for example, like 400, 500 parts per minute, very high-speed line. And imagine people, like, continuously picking up the product from the line, figuring out the defect and putting it into a good or a bad bit, right? It's slow, it's error-prone, it's repetitive, right?

Dhruv Sharma: Just to understand this a little bit better, Aniruddha, is every product passing through QC? Because very often when you look at them, they all have a QC sticker, or do the people on the line sometimes just do random sampling? How does that actually work?

Aniruddha Banerjee (SwitchOn): Yeah, while that depends on the product and the industry, but, you know, for the most common products, random inspection is most common, right? What that means is that once in 15 minutes, you know, someone randomly picks up a product and then takes it through the quality checks, and then they use that to proxy the quality of the other products that went through in that, in those 15 minutes, right? So, this is the most common. There are paradigms in which 100% inspection does take place. Like, for example, if we look at, you know, premium products like electronics, right? Or let's say medical device manufacturing, right? There, it is true that almost 100% of the products go through the inspection. But again, this is a manual inspection, which suffers from all the subjectivity, you know, that a manual inspection encompasses.

Dhruv Sharma: I'd say, I mean, you needn't put like, a subject like average consumer products to 100% quality checks. But if it's like a zero failure kind of product, where it's like a life and death kind of situation, like an escalator or many others, then you have no option but to do this. I think we'll dive deeper into the product suite in just a bit, Aniruddha. But I want to stay with the whole zero defect thing for just another moment. Is zero defect like an aspirational goal? You know, or is it the result everyone is seeking? Like, how many mistakes are people willing to make along the way? So again, the answer is, it depends. But it depends upon, you know, kind of the criticality of the component.

Aniruddha Banerjee (SwitchOn): I think aspirationally speaking, all manufacturing wants to go to zero defect, right? Contextually, what would they let pass, I think in automotive, it's about 20 parts per million, right? 20 to 50 parts per million.

Dhruv Sharma: Define a tolerance limit, and then say fewer and fewer from here.

Aniruddha Banerjee (SwitchOn): That's true. That's true. That's true. And obviously, Toyota being the most famous for this, you know, and many, many other manufacturers following their philosophies, right? Contextually, in consumer goods, for example, this can go up until, you know, about a thousand to two thousand defects acceptable, right? There, by the way, the metric is called defects per million units, DPMU, right? And that can go up to, let's say, a thousand to two thousand defects per million units. So, you know, that's kind of the range people shoot for.

Dhruv Sharma: I mean, what is the range they usually have?

Aniruddha Banerjee (SwitchOn): The range they usually have when we start working with them is in the eight to ten thousand defects per million units range, right? And that's like almost ten times above the acceptable limits, right? And by the time they have installed this kind of AI, vision AI system, you know, they are able to reduce this to a thousand, five hundred. So zero in absolute terms, if you're manufacturing at scale, zero in absolute terms is like, that's a pipe dream, but you can reduce it by an order of magnitude.

Dhruv Sharma: And but now we're going to talk more about the product suite. But before we do, Aniruddha, I'm sure you've read this book, but I think we should totally recommend this book called The Machine That Changed the World to our listeners, even if they're not from manufacturing. It's a book about, so Aniruddha brought up Toyota, it's a book about lean manufacturing, Toyota's production philosophy. It's really good, even if you're not like a hardcore engineer, even if you work on the software side, everyone should read the book. Tell us more about the product suite.

Aniruddha Banerjee (SwitchOn): Yeah, absolutely. So, so the product is a combination of software that goes in all levels of manufacturing. So we start with the inspection that happens on the shop floor. We install camera, light, compute systems, and of course, our DeepInspect software to ultimately inspect products as they get manufactured in the line and then reject the defective products. This enables our customers to reach zero defects after the product gets installed and is running. From there, a lot of our customers want to understand the major root causes of the defects, right? And then ultimately, you know, optimize the manufacturing process itself. For this, we have the software that enables them to classify the defects, to identify manufacturing processes that may be defective. And then from there, we have the DeepInspect cloud that enables them to generate a lot of global analytics, enabling them to compare, for example, the behavior of their SKUs and their sites, right? For example, line one is manufacturing an SKU and line two is manufacturing a different SKU. So you can essentially compare the two lines and all of this in real time, by the way. So the typical way this is done is in a manual, manual review, that is like a bi-weekly visual inspection review.

Dhruv Sharma: Absolutely, right. And now they can, A, do this automatically with AI and B, also use the data to ultimately figure out the bottlenecks that they have in their manufacturing facilities. Aniruddha, do your contracts allow you to, can you name your customers, some of them for us?

Aniruddha Banerjee (SwitchOn): Sure. So we work with very large companies like Unilever, BOT, P&G, you know, and many, many others. These are in the consumer goods, electronics, automotive, and pharma supply chains.

Dhruv Sharma: So I think the next question is going to be on behalf of, you know, a founder who's, you know, in the same space as you. I'm in a similar spot, not the same space literally, but someone who's in manufacturing, maybe a few places behind you and hopes to have those logos, hopes to be serving those logos. Tell us more about the sales process. What have the learnings been over there? How much of it is, how much are the, how much of the learnings are transferable and everything?

Aniruddha Banerjee (SwitchOn): Yeah, absolutely. I think I'll have a mildly hot take here, and then maybe I'll talk a little bit about this, about the, about the steps in the sales process. So I think the first thing, yeah, so the first thing I think is there is this predominant philosophy around, you know, start with an SMB and then go up market, right? I think this is very predominant in SaaS. Particularly for industrial, I don't think that works very well. I think you should straight away sell to the class of manufacturers that you would ultimately want to have in your dream list. That's because, you know, just because of two things I have seen. Firstly, you know, the requirements of an SMB versus an enterprise are very different, right? So, you know, the product and the learnings don't necessarily translate between an SMB and a large enterprise. Then the second thing is that now with AI having such a large tailwind, right, I would say the large enterprises are the ones that are putting a lot of dollars behind pilots and POCs. So it's not difficult to convert them, right? So if you're able to identify a value and you think that is, you know, that is for a customer that is a large enterprise, I'd say go for them, you know, rather than go through this whole upmarket motion, right, of doing things. So that's number one. I think number two, on the sales process bit. So for us, I think what we've generally seen is that we have figured a way to do a very, very quick single day trial of the product that shows the value of the product very quickly to the customers. I believe that this is very, very important. I think a lot of conversation happens around what is the value that our customer gets from our product. And if you can find a way to short circuit that and make it really visible to stakeholders at all levels, it becomes really quick and easy for you to enter, right? The other part also, you know, that you may want to know is that there is like about four or five levels of approvals from people that, you know, sit out of, let's say five to six offices around the globe, right? So, you know, waiting for those approvals to go one by one versus doing a trial where all of them are physically present is like the second one is a much quicker shortcut to, you know, getting new logos. I think that's number one. Number two is, I think if you're looking to sell to large enterprises, you have to have like a growth motion in place from the day one, you know, the amount of time and effort that it really takes to convert any large enterprise means that almost 50% plus of your revenues need to come from expansion right from the year two, right? And I think this is a very underappreciated part of the motion here. So typically what I've seen is a lot of founders focusing on the first sale, but not enough of them focusing on how do you grow that sale, you know, to have a lot of very high NRR, you know, and serve a lot of these customers in a large way. So I think these are the two big takeaways for me, you know, till now.

Dhruv Sharma: I think that is a very, very insightful answer, Aniruddha. And you know, the point you made earlier about decide whether you want to sell to SMBs or large enterprises, I think Sam Altman has a heuristic for this, which is don't do X in order to do Y, because then you need to do X not do Y. So either X or do Y. So this is the same idea in practice.

Aniruddha Banerjee (SwitchOn): Yeah, yeah, no, absolutely. And I think I've read this other thing, which says keep your main thing your main thing.

Dhruv Sharma: Yeah. Right. And I think that's another, yeah, I think the other advice I've gotten a lot, right, maybe this resonates with you. So it's going to take the same amount of blood, sweat, tears, well not blood, but sweat and tears to serve a small client and a large client. So decide, right? Like, I mean, it's, it's going to be the same amount of effort, like, and, you know, SMB companies are great. But like, decide who you want to be. So, so there is that. I did have one sort of follow up to, you know, your specific sales process, Aniruddha, which is when, when it's a, it's still a very early conversation, you've not yet had a chance to even deploy on the line, then, then how do you demonstrate value? Is it from, do you show proof from the previous customers that you worked with? Do you get data ahead of time and do a simulation? Like, how do you, how do you make a case for, for what you're saying?

Aniruddha Banerjee (SwitchOn): Yeah, no, absolutely. So I think, while everyone goes through this figuring out journey, and I think different products work in different ways. But the way our product works is that, you know, it gets, it's actually pretty much plug and play. So you can pretty much deploy the product on the line, and do a, and show a trial within one day, right? That's, that's unique to how we built our product. I highly recommend folks to, you know, design that product in a way that aligns to this philosophy, because, you know, this is the thing that allows you to do a trial on the site. What I've generally seen is that, you know, there is a very, especially in manufacturing, there is a whole lot of variation between different sites and different customers and different standards that they follow. So, so translating a technology that fits very well on one manufacturing setup, may not, you know, may not fly with another manufacturing setup, unless you're able to prove that the manufacturing process is exactly the same, right? You know, and it really is because of a lot of very nuanced changes or differences in the manufacturing setup. So instead, we did the other thing, which is we productized our system to a point where you can do a trial in a single day, no matter how your manufacturing line looks or behaves.

Dhruv Sharma: Right. I think for my next question, I'll first of all ask you, what's your current headcount? Like how many people do you have?

Aniruddha Banerjee (SwitchOn): So we have close to 73 people at the moment.

Dhruv Sharma: 73. Yeah, got it. And then you're working with all of these large customers. So how have you built the org in such a way that you can do, you know, several deployments? Or how do you sequence events? How do you send people on site? How do you spend time personally? How have you built like, you know, who are the other leaders in the org? Give us a sense of all of those things as well.

Aniruddha Banerjee (SwitchOn): Yeah, absolutely. So I think, I think there are a lot of questions here.

Dhruv Sharma: I'll try to maybe I'll summarize, right? Like, how do you avoid capacity from becoming the bottleneck?

Aniruddha Banerjee (SwitchOn): Yeah, yeah, no, absolutely. So I think one of the key things I have learned here is the importance of planning. Right. And I think, I think there are a bunch of very, very good books on this. I think, generally speaking, what I've seen is that, you know, if you are able to visualize what the end state looks like, and the resources that end state would consume, you have to pretty much design all the processes that help build a sustainable and viable business, right? So that's, that's kind of the overall summary of this. How we specifically do this is we specifically firstly try to understand all of the components of top line and bottom line. Right. And from there, we've, you know, based on our understanding of margins and cost structures and so forth, we allocate, you know, different kinds of budgets and capacities for different things. This, for example, for us means a, you know, like a cost of goods, be like a deployment capacity, see, you know, inventory and other kinds of capacities, right. And then what we do generally is that we have, we have a rule of the number of deployments that we do. And that scales up according to our annual operating plan. And we provision, you know, capacity based on that annual operating plan, right. And now we have a seamless process of doing this, because we have a seamless process of doing an annual operating plan. And then, you know, trying to stick to it, or see this itself has happened through a lot of learnings and iteration cycles. But that's where we are right now.

Dhruv Sharma: Right. And did you start your career at NVIDIA?

Aniruddha Banerjee (SwitchOn): That's true. Yeah.

Dhruv Sharma: How cool is that?

Aniruddha Banerjee (SwitchOn): Yeah, it is.

Dhruv Sharma: How did you land there?

Aniruddha Banerjee (SwitchOn): So, so, so I originally started in hardware in a company called LSI, you know, got acquired by Broadcom. And then, you know, for me, the iteration cycle of hardware was too slow. I wanted something that has a quicker feedback, quicker iteration cycle. And then I moved on to software, you know, work for a bit at, at Samsung in India. Right. And then, you know, was a part of a lot of the system software development there. And, you know, got an opportunity at NVIDIA. And obviously, NVIDIA being NVIDIA, I was, I was a part of their automotive software team, which is, by the way, one of the first teams using AI for, you know, various applications, including autonomous cars.

Dhruv Sharma: Yeah. Right.

Aniruddha Banerjee (SwitchOn): And then got an opportunity to work, you know, on one of the chips from the very beginning of it in the wide world to seeing it kind of live and working with customers. So it was an amazing experience for us, for me at NVIDIA. And I think learned a lot from there. And a lot of these learnings I carry into what we are doing.

Dhruv Sharma: So as someone who's ex in NVIDIA, like how at any given point in time, and you'll, of course, remember from your time there, how many different things are going on at NVIDIA?

Aniruddha Banerjee (SwitchOn): Oh, I mean, when I was at NVIDIA, it was still like a small company of 5000 people. So, you know, there's like,

Dhruv Sharma: you were at NVIDIA before it was so cool to be at NVIDIA.

Aniruddha Banerjee (SwitchOn): That's true. That's true. But we built all the cool things. So yeah.

Dhruv Sharma: Yeah. So anyway, my question was, how many things used to be happening at one point in time?

Aniruddha Banerjee (SwitchOn): I think what I've learned of great companies, including NVIDIA is like, there are hundreds of experiments that go on at any time in such a, you know, in a big yet efficient organization like that. Right. And I think, you know, the I'd say the value of doing experiments that fails is rarely understood if you're not in that kind of an environment. But I understood that very, very closely. But you know, to answer your question, just in our team, which was in the autonomous car division, there would be hundreds of experiments we'd be doing, you know, on pretty much everything like this includes the software, you know, go to market experiments, system architecture experiments, software experiments, application experiments, partner experiments, and so forth.

Dhruv Sharma: I think Jensen even has a quote, like, we're paraphrasing what he says, like, you have to, you know, you have to have bad ideas in order to have good ideas. So there is no other way. You know, it's like, people remember Pablo Picasso for like, you know, 10 masterworks, and they fail to realize that he made like 50,000 paintings over the course of a long art career. I think the other thing we don't realize as consumers is how these elite hardware companies, you know, are already have already initiated a product development lifecycle, maybe three years before that, that's true products going to be in our hands. So the insiders know what's coming, you know, even three, four years out, at least a couple of is getting compressed.

Aniruddha Banerjee (SwitchOn): Now, I think now it's more in the range of about one and a half years. Yeah, it's, it's, yeah, I think a bunch of factors, I think there is more customization around the ability to manufacture custom hardware, there is also ADA cycles are compressing. So, you know, there is this ability to write verify, spin out chips more quickly. And so forth. I think there is a there's a bunch of things there. But I think overall, cycles have compressed, but still, yeah, about one to one and a half years at the very least.

Dhruv Sharma: And maybe as the final closing question, at least for today, right, we'd love to have you back is, like, what's what's really happening in vision systems and vision systems plus AI? And how do you plan to take full advantage of that as a switch on?

Aniruddha Banerjee (SwitchOn): Yeah, absolutely. So I think we are at a very interesting inflection point where vision systems have classically been being developed as almost, you know, expert systems with a lot of rules. And what AI is doing at the moment is it is really appending that way of doing things all together. Right. So very, very similar to what has happened in a bunch of other domains. You know, AI here has really come in and brought in a new approach almost to doing vision systems all together, where instead of setting up a bunch of thresholds and rules, and, you know, four to five months to do a setup, you're ultimately able to do a trial in a single day and do an end to end setup within 1015 days. Right. And I think this was already happening. Right. And this is something that we have already taken advantage of. And then what is happening on top of this is a lot of the agentic workflows are really coming in. And they are really enabling vision AI companies to develop the software on top of this, right. So, you know, to be able to root cause the defects to be able to ultimately identify where the manufacturing bottlenecks are, are systems that have now become extremely viable to do. And the development cycles for these have drastically compressed, right. So, you know, you can you can then get this to the customers very, very quickly. The third thing that has happened is a lot of a lot of hardware has become extremely accessible, right. So you now have thermal cameras, hyperspectral cameras, etc, etc. You have a lot of, you know, very high speed GPUs running, you know, deep learning unsupervised models in 50, 60 millisecond kind of turnaround times, right. So, this has led us to really deploying...

Dhruv Sharma: Almost like industrial hardware at consumer price points these days.

Aniruddha Banerjee (SwitchOn): That's true. Absolutely. Yeah, yeah. And that has, that itself has led to so much innovation in this domain. And we have kind of been, you know, very lucky to be on the other side of this happening. It has just enabled us to fulfill a lot more value for our customers, right. And that ultimately has led us to be able to create a much more valuable product. And so these are kind of some of the things that are happening in the Vision AI space. I think, very, very, let me say, very exciting times for us, both from an industry point of view, as well as from an adoption point of view. But I think finally, these technologies are becoming affordable enough for, you know, enterprise to mid market customers to really adopt them at scale. So it has moved beyond the POCs, you know, even with our close to 200 odd deployments that we have, right, it has gone beyond like POCs for most of our customers into real production environments, which is very exciting, very pumped for the next phase for us.

Dhruv Sharma: Is your team already pushing you to start thinking beyond Vision AI and also start thinking of physical AI, like already? Or do you already have some work underway?

Aniruddha Banerjee (SwitchOn): I think, I think physical AI is something that we are extremely excited about. And I think, you know, as you know, AI is, I think, the next frontier of AI is physical AI. So there is like a lot of, lot of things to be done, I think, from a value delivery slash, you know, work to be done point of view. So I think we are very, very deeply focused on the on this problem of taking manufacturing to zero defect. Of course, inside of that, we are doing a lot so that we can ultimately eliminate the cost of quality and ultimately take manufacturing to the aspirational goal of zero defect.

Dhruv Sharma: Yeah. And Vision is going to be a core component of anything in physical AI anyway. So you're very well positioned for, you know, when the time comes.

Aniruddha Banerjee (SwitchOn): Absolutely true. Absolutely true. I think, as you may know, you know, a lot of our, you know, a lot of humanoids are based out of how humans operate and a lot of how humans get inputs is the visual input that we get. It's like one of the very, very key senses for us. So yeah, very well positioned to be able to support that development.

Dhruv Sharma: Yeah. I think it's like with physically, our humans are just relearning how much we have been taking for granted, right? Try building a servo motor that can, you know, spin a ball like, I don't know, like Malinga or

Aniruddha Banerjee (SwitchOn): Yeah, yeah, yeah, for sure. Yeah.

Dhruv Sharma: At a 20 watt thermal envelope. So exactly. But Aniruddha, the pleasure speaking with you. Thank you so much for coming. I have a feeling we'll have you come back on soon again. I mean, either talk about the company or we'd maybe do like a special segment on physical AI or AI in the context of manufacturing, get some of your peer founders and, you know, go a little deeper into some of these topics as well. All the best.

Aniruddha Banerjee (SwitchOn): Thank you so much. Absolutely. Thank you for having me, Dhruv. Have a good day.

Dhruv Sharma: All right, listeners, that was TON 115. I was hosting solo today. This is going to be back later this week. We're not streaming on Wednesday, because that's when we might actually use that time to put together that special on space that I was talking to you about earlier. Well, finish your Monday strong and we'll see you later in the week. Cheers.