250 : Solving the Impossible: Teaching Robots to See | Gokul NA (Founder,CynLr)

Speaker A: If you're too sharp, if you're too smart, you wouldn't plunge into this. Yeah, I think there is some amount of stupidity involved and I think that is important. Gokul is the founder of Signler Deep Tech robotics startup. Signler, or Cybernetics Laboratory, is changing the way robots see and work. Synler creates smart vision systems that help machines recognize and handle objects much like humans do.

Speaker B: How do you iterate in this kind of an environment? Right, because it's not like your typical software fail fast and there's a significant cost to getting it wrong.

Speaker A: You gotta start thinking from, oh, it's going to fail, am I equipped to tackle it? You need to plan your failure.

Speaker B: So where can people seek help?

Speaker A: Tech, I think you're on your own in US Robotics has a very bad reputation. It has not been a very successful investment thus far.

Speaker B: What do you see as the future of humans? Hey Gokul, welcome to the Startup Operator podcast. Thank, uh, you so much for making the time. Thanks.

Speaker A: Thanks Roshan.

Speaker B: Yeah, yeah, it's been a long time coming. I know we had to record a month ago, but I'm uh, happy that it's happening right now. And you have a fantastic office, perhaps the coolest office we've seen possibly.

Speaker A: Thanks. Yeah, yeah.

Speaker B: So, um, I'd like to start at the beginning, right? I mean every business or every startup has this what if? Right? I mean, what if I can deliver food or groceries in 10 minutes? What if I can uh, fit my entire music in my pocket and so on. So what is that what if for

Speaker A: Signlo, I don't have a clickbait, uh, statement to begin with. For us, it's a very, very long journey both for Nicola and I. The bug caught me first. But, um, uh, it was not to this clarity that I have today. Ours is a very slow frog in the warm water, uh, uh, process.

Speaker B: Right.

Speaker A: So, uh, this has been a thought process of what's happening with robotics and automation again, uh, basically a need for an automation starting all the way from my uh, 11th, 12th, around that time. Um, and I was picking engineering and then I picked engineering because I wanted to build these machines. Especially because I come from a, from a borderline town plus village, uh, uh, setup where most of things were already mechanized and automated. But there were these tasks, uh, many of these tasks where you have to do, pick, manipulate sophisticated tasks are not. And labor was very, very um, short in those places. That's also uh, truth in India, it's affordability of automation. That's the problem. On the other Hand that thought process was there for a long period of time and then even during the engineering phase, some of the clarity between what a uh, prototype and as an engineer often we have this hobbyist thinking, right, I'm good at doing something and then, and somehow I'm going to convert this into an organization and so on. So but the thought process is very different when you approach from a hobbyist perspective than to a product, ah, that you need to make someone else benefit out of and should be usable to someone. So with that background, uh, again was um, looking at the maturity of the robotics as a technology. This is back in 2007, eight, nine, around the time.

Speaker B: Right.

Speaker A: Uh, one key problem that we noticed also is the first question that I'm thinking is see the cars have been there for almost maybe 30 years earlier than what uh, robotic arms were. Robotic arms were from 1950s, uh, first arms that came. Why is it not percolating and penetrating into a common life and human uh, life even in industries and businesses or low level businesses like how a truck and tractor has penetrated into a farm. Why haven't they actually penetrated? The biggest problem is volume. And if volume is, why is the volume not clicking in? The more it's not generalized enough, the more it's not useful enough for multiple different utilities. It becomes a problem for you to adopt uh, in volumes. So there is a generalization that is lacking. Uh, then if we dig a little more deeper, these machines are, these are just simple machines. They don't justify the word robot. They are just six motors on a link. And somehow it has always been an undercurrent of an existing technology back in that time. So something market finds as a new tech, they just start coming back and saying hey, can the robots be better right now the fourth phase, our third phase is happening. Three phases of advance of AI and this is the third phase that's actually happening where uh, again we are looking at ChatGPT and OpenAI and then we are thinking whether this could be extended further into making robots more intelligent. Because so those issues were there uh, when what I realized is if I don't have a general purpose robot system, instead of making it very purpose specific to if I have, I have to nut and bolt. If I have to make, I'll be one, uh, there'll be one specialized robot for it for welding, another robot for uh, painting another robot. And in fact if you look at the way industry in itself is not able to adopt robots in large volume only then it'll first come first it has to be Luxury market, then it has to come to your consumer market. And cost sensitive market starts with a niche and then. And then, yeah, and then goes like your Tesla Roadster first and then now you are thinking of Model 3 and so on and so forth. I don't remember the names but uh, the first was a luxury car, right? And that's how everything starts. The industry, which could be profitable, which could offer at a much higher cost in itself was not able to adopt it. Because today if you take 30%, uh, if I take a solution of robot automation, only 30% of the cost goes to the uh, robotic arm. Remaining 70% is all about the environment, presenting the parts. And so, and so customization, customization, heavy customization. You have to control your environment. You have to control how your part comes and mm of orientation difference becomes a problem for robot to go interact with them, pick them, manipulate them. It's a huge problem. Right. Now imagine you have to control the presentation of all the parts that comes your nut and bolt within the space and tolerance of you know, 0.2 mm, right? And size changes. Then whole mechanism changes. M for somehow you realize that no, I can't go and pick it like this, I have to pick it upside down. Then the whole mechanism again changes.

Speaker B: Wow.

Speaker A: Right. With this limitation, how do you automate human being is obviously a better alternative to. And uh, people might argue because if you go to YouTube and then see those videos in, in your uh, you go for you know, automotive manufacturing, you'll see a lot of robots there because those are the only videos that are seen. Right. And often this used to be another question coming from the market saying hey, uh, but isn't everything to be automated? Is already automated, right? What is Elon musclaiming about so many robots and so on so forth. Um, there are two things. One, um, if you look at today, who's the largest employer among the organized sector into blue collar labors.

Speaker B: Labor.

Speaker A: Right. It's essentially manufacturing and automotive manufacturing is one of the largest. Uh, construction and agriculture is a little unorganized sector. Uh though they must be the largest, right? Among the organized sector, they are the largest. So what are these people doing? Right? And if you think about the wages that are paid by organizations like just Ford or $150 billion or $130 billion organization, GM and all of them, they are in like several tens of billions of dollars, right? I think ford is around 7 billion or so. It's just wages alone being paid uh, annually.

Speaker B: Right?

Speaker A: So what is it doing? What are those many people doing? Right. On the blue color shop floors. Essentially, what we forgot is mechanization versus robot autonomy. Right? You still need somebody to operate those missions. M. Or move objects from one location to the other location. That's what people are doing today. Or move tools, interact with the tools, assembling an engine, let's say. Most of those parts are complicated. Imagine the number of wires that go into your car, your roof, uh, all your carpeting inside, and from there to your mobile phone to your, even this mic and your laptop. And all of these are highly unortivitable because M, you cannot control them. How do you control this wire's presence in that particular angle and orientation? Why is that a problem? Because these are just simple machines with joysticks. You tune your joysticks to move it to a particular location, record that position, record to the other. It will just repeat the movement without a deviation from that, uh, movement within, like possibly 0.2 mm to 0.5 mm of, uh, repeatability. Right. But it's completely dumb. Uh, uh, they don't understand what's happening in the environment. And so, and so, fine, putting a camera and then making an, uh, intelligent system, making it intelligent by writing algorithms behind it. That's where the key genesis point. So I'm boiling it all down to this particular point and then finally, okay, how do I make it adaptable? Then you need a sensor that's more holistic to give them a feedback about what's happening in the environment. Now if you put a holistic sensor, vision is the largest holistic sensor, uh, where you can see a whole area of information. Not like a point information. Your touch is a point information. Right? Vision is like million pixels. You're seeing all of them. You can see depth, you can see a lot of information. Unfortunately, we have underestimated the approach of using vision. We always keep associating that with, um, uh, comparing images. We think that if I want to understand an object, if I want to find, let's say there is a pen in this room. It's my favorite pen. Let's say I want to go search for the pen. The assumption that we think how our brain works is it has an image of the pen in its head and goes about searching in the environment. Right? And sorry, starts comparing them and then starts figuring that out. What are we comparing? We are thinking colors for the color for a reflective mirror, finished object or a transparent object. What is color? M. It's the whole reflection of the environment in itself becomes very complex for the system to understand. Right. So this is underestimated we, we human beings have a tendency to usually underestimate some of these processes that we don't understand. While the progress has been phenomenal when it comes to language and language processing, it has been very abysmal when it comes to vision and vision oriented processing. Right. That's because this is the only process that is too natural and instinctive to us. Like to talk. We went to school, right? To read and write. We went to school. Somebody taught us how to eat, somebody taught us how to dress, somebody taught us, possibly held our hand and then made us walk and so on and so forth. Right. But when it comes to vision, you just opened your eyes, it happened. You didn't read a book, how to improve your vision, none of those. Right. So it just happened. Right. And um, the amount of underestimation of processing that's happening in your brain, right. A given time, 55% of your brain is occupied for visual processing.

Speaker B: Wow.

Speaker A: That's why you close your eyes. When you are smelling a rose or let's say or anything, anything good smelling for the case. Right. Or uh, when you want to listen to music or when you're thinking deep or when you're eating an ice cream. The first instinct is to right. So the eye goes closing. Right. So that is uh, a key factor uh, that we are underestimating.

Speaker B: Why?

Speaker A: Because brain has to relieve its processing so that the other senses are in a heightened state. Because m brain consume, the vision consumes. How are we simplifying it? M. How are we oversimplifying it? Have we understood the working of the vision in itself? So those are all large gaps that we realized. And when I was uh, getting into so there was a huge gap in the vision and the vision is so under underestimated. And that's the, that with that notion I joined National Instruments as a, as a company uh, right out of my college. So with vision, uh, focusing on the vision, uh, so both Nikhil and I met there uh, in 2011, uh, at National Instruments. And uh, everybody starts at that particular role and then diversifies into different uh, roles later. And I moved on to become a mission vision specialist, uh, interfacing between the R and D and application uh, engineering team in national uh Instruments. One of the again plaguing problem that the industry had is if you attempt 10 problems, only three would commercially succeed.

Speaker B: Right.

Speaker A: That's the machine vision's reputation. Customers would say hey, if you're solving something with, with uh, vision, please don't. If you have something that you could do with mechanical where I would rather control my environment than to put a vision on top of my robot and then having one person stand all the time to see whether it is going to work this time or next time or when it's not going to work.

Speaker B: Right.

Speaker A: That's again also boiled down when we did the soul searching, why this is happening. And it went to the fact that we are again assuming manipulation from identification. What is the foundation fallacy? This template picture, if you're not able to match, you can't go act on this object. Right. If I pull out a random object of my like let's say my keychain that you've never seen before or crush to paper and I give it to you, it's a completely random object. Right. You would have never had the mental picture. There's no template. There's no template yet you will be able to go pick it. You're able to use your vision, guide yourself, your, your, your faculties, which is hand and whatnot. Focus on that object. Go pick it. Which a baby can. Baby may not know you're feeding a pen, right? You're giving it a pen, it'll pick it, put it in his mouth, throw it out.

Speaker B: Right.

Speaker A: So these are instinctual aspects of how it interprets the world which is absent in vision.

Speaker B: Right.

Speaker A: Uh, what we call as sentience before intelligence. I mean people think sentience is emotions and then a higher elevated stage of intelligence. But it's more about the more primal instincts that the system has. Like if you touch a fire, your body understands sometimes. Ah, if a object, I mean may not be like snake, but something else that we instinctively, uh, babies do react to. And those are structures that are built in your brain which is used to learn and interpret the world. That is a large gap that the machines have today to be able to interact with the physical world, extract the uh, the intelligence about or uh, not intelligence, the information about the world and the objects around and um, those missing pieces is what we realized is a distinction for a human being to be able to manipulate really well. And identification is a unintended uh, uh, consequence of the brain wanting to manipulate the world around it. It's not that we used identification to be able to become manipulation, savvy. We did manipulation of space, either navigating whatnot, then identification became uh. Because there are animals which can, cannot remember yet they can go act on the. To the extent that we can.

Speaker B: Right.

Speaker A: But it can act on items around it, the environment around it. Right. So this gap is what we noticed. And then we went out in 20 uh 15 we left National Instruments, both of us. Um, um, Nikhil was specializing more into the business side of uh, more into the sales side of uh, the account management at National Instruments. And while I was focusing on the vision side of things and building these architectures and teams and so, and so um, then in 2015 we stepped out to test our thesis. So we took uh, with the blessings of National Instruments we took some 30 uh, different failed projects and problems of the past. And then after we had successfully cleared all of them we had 100% success rate. In a space where you had 70 to 30% uh, ratio of success failure. Uh, we just uh, took this as a precedence to go race and then productize this because we were dependent, hyper dependent on the off the shelf components. Uh, it was not packaged and form factored into the way where the vision in itself, the camera itself. We don't have an off the shelf camera that can give you all these abilities limiting. So we started building, we started looking at what does the customer want? He wants a complete automation solution which needs a plethora of different manipulation capabilities, sensors, grippers and vision. And vision was the foundation that you had to begin with. So we started with building that from the hardware in itself. And uh, from 2019 onwards we have been building. That has been the journey thus far.

Speaker B: Fantastic. So if I understand correctly, machines have become very good at very specific purposes. Right? Uh, but what you're trying to do is make it adaptable enough for a wide variety of things that it can do wide variety of tasks and operations and so on. Uh, it seems like such a fundamental thing. Right? Why do you think there hasn't been much development on this for the last 40, 50 years? Or do you think that people have taken a different view of what uh, they could get the machines and robots to do?

Speaker A: I think there are multiple parameters and factors when uh, there are several parameters. So one, how the industry approaches in itself. There's an industry angle to it. I mean everything boils down to people's perception and different departments understanding it in a different way. Right. Of businesses understanding uh, it in a different way. There is a fallacy of confusing with integration and uh, developing something fundamental. Robotics had always been seen as an application domain than a domain that needs fundamental research. M. We are not noticing that it's a component level problem. You're not um, realizing that it's a technology level problem and something has been not working for more than 40 to 60 years now that you are attempting. The first camera on top of Robotica was tried in 70s.

Speaker B: Right.

Speaker A: Unlock camera, you don't have unlock processing, you don't have digital computers. And none of this, um, I'm not able to find this video again. But it was some MIT lab attending that back in the time.

Speaker B: Right.

Speaker A: What's missing is we always thought as it's an outcome of disparate set of technologies that is to you could put them together and make it work. Right. We didn't think it needs its own dedicated set of technology by itself. That's a mistake.

Speaker B: Okay.

Speaker A: And it has always been thought as somebody's incapability of engineering with all the existing tech. Oh, it's possible. And then, and foundationally, vision is very, very, uh, elusive. Right. Uh, so that's why it's elusive also.

Speaker B: Right.

Speaker A: So, uh, we do not understand how the process of vision evolves. So we underestimated, kept underestimating, kept underestimating. It goes on even now. The problem is the moment you see a picture on an image or a video, we are able to understand it really well. Right. We assume that that's what your brain is also, uh, seeing. There is this personification issue. Anything that moves like a human being, we think it has the intelligence like a human being. The moment we see those pictures, we think that we understand. So the machine must be understanding or it's able to break. So if I do some kind of filter, you're able to see the edges and contours, it's able to extract those edges. I'm able to recognize that it's a human being. So it means machine must be interpreting it. It's possible to make the machine understand that it's a human being. Those fallacies makes us, uh, underestimate the nuances that are there, uh, um, in solving this problem. We are not starting from the problem. Many times it's engineers who have been very savvy with particular tools and toolkits and, and when I say toolkits, even subjects and disciplines, right. They're coming to solve this from that front. They're not coming from the problem. If an object is so hard and they have been attempting for 40 years, it's something about the object is a problem, isn't it neon with tech. So what am I not understanding about the problem? Right. This is one, this is on the technology side, on the business side. On the other hand, it has always been an extension of another industry. It's an undercurrent. Technology undercurrent.

Speaker B: Right.

Speaker A: Um, where, oh, this should be extension of some automation of so and so. This should be an extension of CNC machines or this should be extension of you know, some process production automation. And so that again, underestimation was a major problem in the way they tried to commoditize, hyper commoditize very quickly. Not looking at this as a vertical integration problem. Got it. Right. And you're borrowing, you are trying to solve the proofs of market through borrowed components from different, uh, components of different products which were purpose built for some other product. Trying to integrate it all together to make it fit into this fashion. So these are the key issues, uh, why this problem has been eluding the market for a long period of time.

Speaker B: The end impact of what you're building seems to be like a universal factory. Right. I mean uh, I don't need a different PC or a Mac to code, uh, a consumer software versus an enterprise SaaS software. Right. But I certainly need a different set of machine components to build, uh, um, object A versus object B. Um, right. But if you build machines to the level that they're extremely adaptable, then theoretically then I just need a bunch of these machines and I can produce anything with that.

Speaker A: Yeah, yeah. A bit of an insight into this is, you know, some factories, what we imagine could become. Right. That's, that's a visualization that we have or the vision that we have uh, towards what this technology can give birth to.

Speaker B: Right.

Speaker A: On the other hand, the foundation will come from an object computer principle. Right. What we. Which is basically a borrowed imagination. Yeah, it's a borrowed imagination. Um, you can think of an analogical process of how electronics evolved and compute platforms evolved. 1960s and 70s, even before that, the automation was there for a long period of time. People used to do the computing through gears and even before it was pneumatic and so on so forth and vacuum tubes came. Then people started doing punching cards as inputs to those systems and things. But every point of time, whenever there is a customization, people used to have a lot of different missions, custom built and custom circuits and so on and so forth. Right. They were not specific computers built. There are computers that were built for specific processing. Right. Um, in fact, uh, you would have famously seen this picture where in 19 uh, 80s there are so many devices on your desktop which has just shrunk down now to uh, a computer and now to a uh, mobile phone altogether. Right. So it all became this revolution and transformation happened when they, when they realized that hardware has to be standardized, software is still not standardized. Right. What you could do is you can have a Swiss army knife of a device which is equipped with Am I 247 using the webcam there? No. Am I using the microphone 24, 7? No. But whenever you need that capability is there and you are able to bring that within the price point where the consumer is able to use it. Because it's now the same device form factor is usable for so many different varieties. It comes back to the problem. What I said is volume, right? It is. It is inbuilt with the capabilities to do any kind of tasks so the customer can buy and then later decide how I want to use it. And it universally scales and as long as it's commercially viable it's not a problem.

Speaker B: Right.

Speaker A: So bringing them to the commercial viability is a very different problem. It's a volume problem. Right. Utility to be universal is how you. You create an under denominator of or the common denominator of problems. If you don't do that you can't productize. Right. So what laptops brought in terms of hardware standardization is what we try to bring in terms of robots today it could be again it can go into the thought process of what all does it need as a bare minimum standardization that we can get into which is modular enough where today I could use it for nut and bolt, tomorrow I could use it for wiring something. Day after tomorrow I could use it possibly for stitching a cloth. Today we can't. It's a very complicated problem to solve. But eventual goal is to get there that you have one underlying hardware. The moment you have one underlying hardware making it do something else is virtual. It's on your software front right now that's the uh. That's the journey that we want to get into now if you're able to do that, why is a factory which is uh. Now if you think about to make a car to design a car or design a factory, which one consumes the maximum amount of money for a customer today? Right. Factory. Shouldn't that be the. The product that he actually owns than the producer itself but he's bottlenecked to buy the product in itself. Right? Because this infrastructure that you have produced is specific to only that product. So which comes with such a large huge amount of business risk that someone else evolves and then has an advantage to look at his mistakes and then evolve into a new business. He is not able to compete along with them. Because I have to make a recovery of this uh, investment capital investment that has gone into this business.

Speaker B: Right.

Speaker A: We could say that as industry keeps thinking that as a constraint to solve we are thinking that could be a problem that is solvable. M. The constraint is oh, you can't have non specialized machinery. The machinery is what it boils down to the machinery and the process to operate the machinery. If I could generalize the machinery, the machinery can adapt itself to various different uh, tasks. The sequence of whatever I did here could be for another microphone that someone else sitting somewhere else could actually teach and train that robot. And this robot here will transform itself to immediately do something else, some other recipe of task. So when this happens, the factory could scale itself to a variety of different uh, product. Because product is nothing but same aluminium becomes this uh, stand. The same aluminum becomes that robot. Right. Or a car. Right. Or another stand. Right. The difference is how the set of procedures that you did to change them. Right.

Speaker B: So rather than, you know, optimize the input parameters, make it cheaper, faster, better, uh, can I increase the outcomes, vary the outcome in such a way that with the same setup, I mean I can do many different things.

Speaker A: Yeah. And so it's a matter of time and investment back into the component level, uh, to the component that goes into the arms to make them more and more efficient and so on, so forth. Right. So that'll be the Moore's law for electronics that we can, I don't know, I'm just guessing that there will be a law like Moore's Law that will come uh, for it.

Speaker B: Uh, when you start with such a, like a huge ambition. Right. I mean it's really like out there. What does day one look like? I mean what is the first problem you pick to solve? Um. Right. And how do you get your team to kind of focus on that with drawing a straight line to that envision that you have?

Speaker A: Yeah. I think there is some amount of stupidity involved. Uh huh. In uh, in thinking that you could solve it. And I think that is important if you're too sharp, if you're too smart.

Speaker B: Yeah.

Speaker A: You wouldn't plunge uh, into this. Right. There is some amount of element of hope. And plus the question of day one, I think like I said, it's a, for me at least it was a, it was a uh, gradual evolution of industry. Reflecting on you. You're reflecting on the industry. Right. So it's been a gradual growth. Right. And then we had a long term picture. We do not. That's like, you know, you have a peak, something that you want to. The Ketu that you want to climb. You have no path laid. It's like a forest. Right. So we have to group ourselves, find a bunch of people who are exploratory Go a set of milestones and then come back again, regroup, say which is the path that you could move forward and move forward with keeping that in the focus because uh, that's a long both in time and the way and the journey is unpredictable and very long. The first structure is for us to understand what is the first milestone that we can achieve that is visible to us from the viewpoint of vantage point of where we are. So blending along with the industry is uh a primary uh direction set up for you and we have to start what is it that we can actually start today that the customer could be profitable with. Right. And then start evolving and then moving further and further away with right. So uh, of course I would not have a maturity of the technology on the day one because it's, it's first of all foundational technology which means proving it to a product form factor in itself will take a lot of time and you don't have support of an ecosystem. Your price points are not going to be mature your talent, uh, uh the level of talent that you have and the quantum of talent quantity of talent that you have will also be very limited uh for it to scale and so on so forth. So often the interpretations of commoditized market will not work out. If you take the template models of how Swiggy or Avola or such businesses which are business model geniuses, uh which learn to leverage the infrastructure maturity that is there.

Speaker B: Right.

Speaker A: For us we have to start with infrastructure in itself. That's the first thing that we stare at right. Uh as a challenge. Right. Second, industry will not provide you with talent because this industry is non existent today and what you're building is not a company. You have to build an industry. If you don't have that consciously in your head then you are underestimating the problem. Right. So with all these parameters to play you start looking at where all my advantages in lead way uh, would be.

Speaker B: Right.

Speaker A: And you have to search for those hunt. There's a lot of, it's a lot of experimental uh progression that is involved in this. So so clearly to. So we are staring at a lot of unknowns. Only thing for us to see is where is the first hook for us to hang on to to progress further.

Speaker B: Right.

Speaker A: Customer is one, vendor is the other. So you have to balance between them strongly. Investor would be the third investor would react to this hooks being available. Uh that gives them at least a modicum of trust to say hey you could move forward. There is something that we can go for the next step and the Next step. So I think that's how this market even for a well less business model innovation, it's the same case. We have to start with the infrastructure innovation in itself. So it's even more harder, even more riskier. And then as we keep grinding, an industry starts slowly forming around us. Uh, and the cycle thus far, that's how it has happened.

Speaker B: So you find a manufacturing service that you can offer, productize that and then build that capability uh, further up. Right.

Speaker A: Essentially more than the manufacturing service. I wouldn't put myself in uh, a

Speaker B: limit of the scope of problems and then look for people who are going to pay for you to solve that

Speaker A: problem, build that thing for them and

Speaker B: then develop the technology and build on that innovation to make it sufficiently productized that you can now take that to someone else, their peer perhaps and say that look, we've solved.

Speaker A: There is one mistake that typically everybody does. Uh, many system integrators who wanted to be a product company ended up being only a system integrator because you think that if you supply, if you do it for one customer, it's, it will magically. The product versus system integration thinking is confused in many cases.

Speaker B: Right.

Speaker A: Uh, now, uh, but a step back, uh, onto your point, just resonating more onto your point. It's true we need a customer, but the customer is not going to define the product for you. What do you have as a technology? You have a know how that you have invented saying that if you have a camera like this, if you have a robot like this, if it has these capabilities, it has these features, you would be able to solve the particular problem. But would you be able to fit into the price point? Would you be able to fit into the, you know, necessities of the, of the customer, their processes and then how it interacts with the rest of the other components in their industry. Right. Often you are, you are. What you have to realize is you don't have the convenience of B2C. Where you are you yourself in itself is complex.

Speaker B: Right?

Speaker A: Um, because you yourself being a ah, customer, you don't have that luxury here. Your customer. Your problem is not that robot's incapability to put two parts together. Your problem is that business, if m. You don't understand whether the business can be profitable around this, then you don't have a organization around this. You don't have a possibility of a startup or a company around this. What you're working out is the business problem there. You have to trace that back. Right? And then you need to take, if you Ask the customers. They'll always tell you. Faster horses. Right. Ford statement. Right. Uh, uh, that's because they think their solutions through what is available, what they are sure of. You are taking a technology that is not yet proven.

Speaker B: Right?

Speaker A: Right. The only way by which they are going to give it to you is when you take. When you own the risk of his final utility, which is the whole solution. He's not going to bite on to saying that oh he's built a new tech. I hope that this is going to work. I'm very, very confident that this is going to work because his neck is at the end of the day within his organization. It's an organization, it's not a consumer market where he can take his own uh calls. He's answerable to so many other things and then it could cripple the cycle of the organization, operational cycle of the organization and so on and so forth. So the system is built very strongly to ensure that mistakes don't happen. So the only way by which you could do is you need to take the risk, you need to borrow the risk. The more you mitigate the risk for the decision maker, the sooner he is going to. That's the reason why you have to go with a solution Intel M and you have to prove it to him how it will work. Then you have the next phase of work where you have to automate this through the infrastructure like system integrators and uh, other suppliers and distributors that you might have as a business chain to take it forward. And for that you have to further productize your training, your productize your support, productize your. What do I mean by support and productization? It should be same. You should need to, you need to a certain extent templatize it. Otherwise you can't scale. M Right. The scale thinking is what is missing often when they go about with solutions. Many do figure it out, right. Whether they articulate it this way or not, they understand this and they go with solutions. In this industry predominantly they go with solutions. What happens is how do you productize it? M so that's where the slip happens. Often they leave that to market to dictate it for you, you need to constantly make effort. Unfortunately this model is not very VC savvy or VC exposed and VCs are not exposed to this model to make it work. It's a long gestation things neither their fund thesis allows you to do that because it consumes a lot of cost. Not only you're investing in tech, then you are investing into solution. Then you're investing into productization, then investing into an ecosystem, then you have to invest into your support network and post sales support and all that.

Speaker B: Right.

Speaker A: It's a, it's a, the payoff is, payoff is longer and longer gestation period which the fund in itself may not, may not allow. So you need a VC industry with layers of passing on and packaging and then moving on to the next because the LPs may not have the, have the patience for that.

Speaker B: Right.

Speaker A: But you should be able to package it and then give it to somebody else and then they should be able to take it up and they don't have a benchmark and milestones for them to understand how to uh, judge this in these cases that metrics are missing. So all of this has to be innovated from the trend. That's what will take you. So these aspects that actually comes and we just think only from a solutions point of view, your product needs to inculcate all this, uh, the metrics of the industry. Right.

Speaker B: So yeah, now we spoke about how you have to pick a solution, Right. Or uh, rather you have to deliver a solution and along the way you have to kind of productize it and make it generally applicable for let's say the industry at large. For let's say someone who is at that phase, they've identified an interesting uh, area to solve in. Um, how do they source the right problem, how do they work with the right customers, the vendors, um, how do they go from that idea to product to company? If you have some principles that you could share. Sure.

Speaker A: I mean when you're saying about idea, um, I still need a distinction whether it is a technical idea that they have or is it the business model idea that they have.

Speaker B: Let's say like a broad technical idea. Right. Someone says that they want to do something in computer vision, for example.

Speaker A: Sure.

Speaker B: Yeah.

Speaker A: Perfect.

Speaker B: Yeah.

Speaker A: So one of the, again I think I'll again pull back the point that I was saying many times really skilled engineers start thinking they could productize their skill. When you productize their skill you could just become a service company.

Speaker B: Right.

Speaker A: If you have that clarity, then it's great. If you want to just say that I'm going to build something and it's going to naturally become a product that jump and you know, that assumption, that visual thinking that we get into, uh, like Elon Musk says, that's, that's. We all of us had the fallacy, right. We compared to my crowd around me, I have this, an edge and I see that somewhere this could be Applied, which is going to be a big problem. And then you have perceived that a super large problem when you yourself understand the economics of what it actually costs. Because you know you yourself are a user in that case. Right. So which is not a luxury that you have when you are in a B2B. When you are moving to a B2B business. Right. Some might originate from within the business. Some might originate as a somebody who's just stepping fresh into the industry. Right. Um, I don't know how you will get those perspectives. You could work for an industry, you could work within that industry. Uh, what not right. But you need to understand that you are not solving that technical problem for them. You are solving a business problem. At the end of the day, it's a B2B business. You are talking to a system, not talking to an individual. Right. Uh, if we don't make the distinction and then plan ourselves accordingly to that and then map out the nature of your problem, which is your customer, your customer is not that individual. Your user might be an individual somewhere within that industry, but it has to pass through a whole system. The benefit, the sponsor, the again, someone who's going to monetize on top of that is all a system.

Speaker B: M Right.

Speaker A: So understanding what is the what, how does it translate to a pain point to the system is where you need to start thinking about. And second, is the system aware of that problem? I have to go teach the system about the problem then. That's a huge, huge, uh, barrier you are looking at.

Speaker B: They may not be solution aware, but they have to be problem aware.

Speaker A: Problem aware. I'm talking about problem aware. Right? Problem aware. See this also we face, right? So they are aware of the problem, but they believe that this is not solvable today. So they deprioritize it. You need to know why do they think it is not solvable today? I'm not saying. I mean every disruption is breaking the thought for the customer, right? Otherwise they are going with a person with a hammer, always thinks with hammer, right? He will cut the tomato also with a hammer. We'll try to do that. That tendency is there for all of us. The industry also carries, of course, it's all human beings and they will carry that thought process. System also has that memory in itself. If you're breaking that, you need to know why is that perception? What did he try before? Why did he fail in those cases? Why were they so negative about. For us it was visionary. The moment you say with vision, they'll be like, no, I know we have tried this morning to evening lighting changes. It doesn't work, right. It's not eliminating a person for me. I have to put a person there and at least if it's periodic, it's fine. For every 10 minutes, one for sure. I needed a person to come in there. But if I do it for 10 systems, I am replacing 10 people with three people. Possibly. I'm just giving an example. Then at least seven people reduction is there. M. So okay, fine, that's a system that I can adopt. But you don't know whether it is one minute once you need it or 10 minutes once. So you need to have that person continuously there. So erratic need for availability of people. If you still have your technology, still demands that that's not going to work. M. I'm just giving one insight like that. There'll be several other things, right? How does it roll down to his supply chain? How does it roll down to his sales cycle? Uh, what is the benefit that he's going to get by doing this? How is he going to push? Where are the organization goals set? How is the budget system working within the organization? And sometimes as bad as the person who was sponsoring this inside the organization moves out, right. You're gone. Things like that. Then how do you insulate yourself from those risks? How do you paralyze to have multiple and one customer here again for our industry, I'm saying in a B2B, um, one customer, uh, is multiple accounts or one account is multiple customers. Sorry, one account is multiple customers.

Speaker B: Right.

Speaker A: You might have a label of uh, let's say somebody like uh, automotive giant like GM or Toyota or Ford or somebody within them. There will be multiple different customer ranges and so many departments, divisions and across them. Right. And again, you need to know what layer you are selling to. If you are somebody who is selling a wire and a cable, you are selling at the plant engineer level. If you are talking about a process and a transformation, you are talking to somebody else. If you are talking about business model modification, then you are talking to the cfo. So you have different customers within your. Whose problem are you solving? It could be technology, it could be the same vision, right. Is it a QE problem that you're solving?

Speaker B: You need to understand procurement. Basically.

Speaker A: You need to understand, you need to understand structure of the business.

Speaker B: Right. Because it's a well considered, uh, decision to use one thing or another. It's not like an impulse purchase.

Speaker A: It's not an impulse purchase. Right?

Speaker B: Yeah. So you need to understand what are the, you know, the com. The complete hub and Sport of all of this stuff.

Speaker A: Correct.

Speaker B: And who you're selling to.

Speaker A: Correct. And you also need to understand how much is their experimental budget. That's easier said than done. Uh, it's harder. They will not reveal. But you need to have. So sometimes having that inside leg allows. M Right. And you'll also have a problem of how would you sell your tech to the. You need the right ears, right person, you need to go to the right. You can't just interfere with the purchase guy and then hope that this will, um, work inwards. Who's the end customer within the organization? Right. And who realizes this very clearly.

Speaker B: Right.

Speaker A: So this comes back to the again, the problem of have they realized that? Do they realize that it is a solvable problem or do they realize that it is an unsolvable problem? If they think that it's a solvable problem, you don't have much scope. Most probably they're a popular resident. You will be one among the competition. M if you're okay. So even there, are you trying to compete with someone else because you have a better tech or are you competing with, uh, you know, an unsolved problem? M so that also you should be very clear about. Right. So we looked at for us, we looked at that it's a fundamental technology. It took a lot more time than what we anticipated. Uh, or kind of we were prepared for it, but what the stakeholders were expecting out of us. Right. But uh, it is it for us. It's a fundamental technology which means we can't be talking to the system that is meant for procuring competitive technologies that are already available. You need to talk for somebody who's there within the organization, who's thinking of future transformations, who's thinking five years ahead? So that goes to the leadership all the way up. It has to trickle top down. Is it bottom up or a top down problem? Right. Some things, if you go top down will not work. You need to go bottom up something. You have to hit both. So you need to know, where do you fall the problem that you're taking. Where do you fall? Right. So that systemic understanding of the customer and the industry space that you're dealing with. And by the way, each industry will behave differently. I can say manufacturing, but the manufacturing of food is very different from manufacturing for automotive, automotive tire, one oem, um, versus your electronics. Each of them have a different process of different setups, different layers of decision making.

Speaker B: So does it make sense to first narrow down what field I want to solve for and then keep them parallel? Yeah.

Speaker A: Keep them Parallel. You can't leave out your technical advantage. You have to, because that itself consumes a lot of effort and time and understanding the foundations and fundamentals of what you are actually building. While that is at it, don't sequentialize this. Keep this parallel. You need to also go understand the problem, don't you? You, as much as you are excited by understanding the tech and your ability to apply the tech, you don't put that much energy on knowing the problem, spend that as part of the problem and understand, think that this, this whole structure is the problem.

Speaker B: Right. This is a typical engineer's mindset.

Speaker A: This is a typical engineer's mindset. Right, right. Which I am also, you know, uh, at blame for.

Speaker B: So, like one, follow up on this and then we'll move on to some of the technical challenges that you've solved. Right. At signlow, uh, how do you find the most profitable problems to solve?

Speaker A: Interesting. How do I find the most profitable problem to solve?

Speaker B: Let's say, I mean, you have four, five, seven years in the industry. Uh, you have a certain skill set. Now you want to be able to map this skill set to a problem that the industry is paying top down for.

Speaker A: Sure. Yeah, yeah. No, I don't approach that way. At least I didn't approach that way. I'm not saying that's wrong or right. I'm not so good at competing because, uh, it's a distraction for me. Um, I would. And I don't go with. So anything that peaks in its, uh, this is a challenge. Right. Anything peaks, peaks at its remuneration, has had an exponential growth will also have a dip because that's the first thing the industry would want to replace. Right. Volume is very different from profitability, which means it's expensive for someone. That's a profit that the customer can eat. Right?

Speaker B: Right.

Speaker A: You have to borrow that for you. If you have margins exceptionally very high.

Speaker B: Right.

Speaker A: It either evokes more competition within your space, which means you are hitting the saturation point. It's going to start henceforth it's going to get hyper commoditized. That's why you're getting into. So the business will have a transformation. You are at the edge of the business transformation. Right. Like starting Swiggy or Uber in 2000s. Early 2000s is very different from starting in 2011 and 12s and 13s and so. And so the business is. Difficulties of the business is very different. Right. The journey that Flipkart went through in India or Amazon went through in globally. Right. We always talk about the fast mover advantage. And uh, there is a lot more of fast mover disadvantage because you build a whole ecosystem around you that's never talked about. And, and most of the products that you use is not built by the first guys, including your Google and uh, so on and so forth. Right. So um, so profitability is something that you look in terms of size of the market that you can capture than I look at it from that point of view than the depth of margin that you can get. So you can think about I will sell only 100 pieces every year but I can have a large depth of profit. That's a car niche area where um, where they are willing to do premium. So if I work backwards from what is that you could replace at the same cost as what the customer has today of which you could make maximum profits. So that's something that uh, we should kind of uh, trace back in our customer uh journey. So if I start looking at is this the product that I can maximum, uh, maximum earn where the margins are very high. Like if, if, if I am replacing a human being, what kind of human being am I replacing?

Speaker B: Right.

Speaker A: What kind of value am I replacing? What kind of labor am I replacing? What's your charges? So specialized work or is it a. Like for example there are tasks that needs like 8 shifts versus 3 shifts. M where it's, it's, it's ergonomically a uh, huge issue. They can't underbelly of a car. You can't just keep doing all through the day. So maximum two hours right after that because it's a continuously moving car and then you are continuously operating on it. You have to keep moving like a, like a honeybee.

Speaker B: Right.

Speaker A: You are continuously on it. Right. So after two hours he can't sustain doing that. Then another person comes. So the cost for doing that activity is very high. M for the customer. That's something that you borrow and then automate first. That's a luxury second thing. I also look at initial days you might want to have. You will not be able to make your margins in spite of you being exponentially expensive. Why? So your supply chain itself is not uh, mature. Your volumes are not mature. I will get at a premium compared to someone else, some other guy, the same components, some other guy who's actually give them a volume business. So my net cost of the solution is very high. Yet the customer is supposed to be profitable on and above that. So you need to look at the roadsters of your uh, within your. Which is the luxury thing. Often initial Picks like, and you need a patient customer who's okay with this failing. Often that will be somebody who's trying to build an image through your product. And it's a very high technology product. Right. So they want to showcase this and they want to add the value of their branding more. So that's another way you could tap initially. Right. So otherwise, um, I am not very clear about where will you get your high margins and high profitability systems within that. That's more of your business model working internally than. So how we started with is we fixated the price point. This is what the customer is willing to pay, this is what they can afford. And then we worked backwards from there and that was a conservative amount. By the time we caught up, the bare minimum has actually rise, uh, has grown, uh, higher in the value because of COVID post Covid and supply chain issues and so on, so forth. So sometimes the serendipity also works, but otherwise you just look at how much are you consolidating as expenses and then you could actually take that away. The second way to think is

Speaker B: often

Speaker A: if you can get into services added on top of your hardware platform, that's a great value addition that you can start looking at. Right? So that's a good margin business for you, profitable business for you. So I wouldn't think about, hey, hitting the price point right is a big, uh, problem. Right? So can we get the price point right? Can we get the price point right? What if I have underpriced or what if I have overpriced? That is a bare minimum price that you have to work as a model and you have to start looking at. Let your customer set the upper price.

Speaker B: Got it.

Speaker A: Keep your services open so that you can extend and then later you can just keep adding that more and more and more and more into this.

Speaker B: So when I said the most profitable ideas, I didn't mean it in the precise sense of the word, but this was an education for sure. Right? Um, yeah, I mean, since you talk about price. Right. I mean there's, there's technical innovation and then manufacturing is a whole other thing. Right. Uh, Elon Musk himself, uh, mentioned this, that, you know, manufacturing is a, is

Speaker A: a whole other beast. Yeah.

Speaker B: It's a different beast altogether. Right. So, um, how does one understand the nuances of manufacturing enough to take product to market? Right. I mean, um, whether it is interfacing with vendors, whether it is like picking what materials you're going to use. Um, yeah, the whole lot of it.

Speaker A: Yeah.

Speaker B: Um, we could spend a Couple of hours on that. But you know, if you could just like cover the broad contours of it.

Speaker A: So manufacturability, um, I think again one um, fundamental question that you have to ask because it depends upon where is the background that you're coming from. If person who's within the industry. Even when I was within the industry I didn't understand manufacturability to this extent that I understand today. That's something you have to bite a little bit.

Speaker B: You don't understand components and so on

Speaker A: because uh, usually even within those industries these are marked different roles that are kept and there's a whole team behind it and then comes around that. Right. So but the moment you start understanding the pain of transactions and how hard it is to you just think that oh, I pay something that I'm going to get, you won't get it.

Speaker B: Right.

Speaker A: Like one of the first experience that I had is um, the first robotic arm that we bought from one of the vendors from within the country. Uh, very initially in 2019, around uh, after we raised our funding in August the second half of the year, um, around October, we've uh, you know, placed the order. The lead time is at least you know, around 90 days or so. And that person got the order from us and then he left the company. Right, left that uh, the, the, the vendor's company and he just shipped the wrong product to us. It's a wrong sale.

Speaker B: Wow.

Speaker A: Right. And that time it's such a small amount. Right. It's 770k and one robot setup in itself costs you around 300k a lot of times. By the way, there's one more point that I had to kind of um, highlight many times you would want to. When you're dealing with this complicated industries, you want to start with your own money, try to do that. And so, and so it's possible in your software because your investment is just your laptop. And then you could people and laptop and people could be your co founders or it could be your uh, founding member teams and even yourself for that case. Right? That's not the case. When you come to a proper industrial gadget and things and your trials that you show their customers not even look at it, they need a system that can work for next five years. So you need to begin with something that you can off the shelf which is proven for working for five years and so on so forth. But when you have to deal with all these complexities, how you could work uh, with those variations that can happen, then you will know how Much of effort that is needed, how the thinking should be, what we should do around now, you will see that there is time as a factor that comes in. So just because you pay the money, you're not going to get that. What are you paying the money for? How much cost, how much of premium you are paying for each of this hardware. And you also need to understand that uh, those guys, this is on the supply chain, not I'm talking about not into the manufacturing, right? Um, when you are paying this premium, you always remember to think that he's paying the premium. Where is his premium coming from? Is he trying to earn most out of you? You're a startup, they should have given you a discount. All this you can think about. But for him it costs more. Because a TVS who might buy 50 robots versus you who's buying one robot. Both of you would need one application engineer full time being kept there. But the cost of that application Engineer shared between 50 robots for you, it is shared between one robot. Even if he doesn't get a single robot sale for next six, six months, he's not going to accommodate you. Even though you're giving, you know, you're saying that I'm going to give him one. One robot possibly every three months. Once is not going to accommodate you. You'll rather take that sales time and effort and then apply for a customer who would have a potential tvs. M may not buy a 50 robots upfront, but he would buy one robot. But the one robot that there is an assured potential of 50 translating into 50. That's not going to happen for you, right? You need to think of vendors as your investors,

Speaker B: right?

Speaker A: Every stakeholder essentially becomes an investor from the front, right. Somewhere somebody should be willing to invest that a's extra cost. He has budgeted in his margin saying that I'm going to give only these many hours of application engineering or post sales services or sales effort for you. But when I am interfacing with the customer, the moment he sees, I mean the vendor, the moment he sees the email that comes from me, you'll be like this is not. He's asking so many things. Everything he's asking to leave alone after that. You asking for customization, right? This is on one side. Next if you are bringing to the.

Speaker B: There is.

Speaker A: There is an angle of price point that you can achieve the cost of manufacturing which is the number of hours that will take for you to make that particular product. It's not important enough that you have made something that works, right? Customer has certain benchmark Of a price. You have your BOM cost already. There is a. There is a leftover of margin within the money out of which you have to run your company which includes your R D licenses, your products, capital, sometimes capital investments, but predominantly operational cash cycle that you have to kind of make it work. Now imagine the. Imagine the layers of post sales pre sales interaction that you're doing too. And plus how many people can produce are uh, needed to produce their own particular product. How many hands is it going through? How many manners does it require for you to produce this product if you don't include that into your. This is applicable only for hardware design software is all this complexity of business is kind of gone in your software side of things.

Speaker B: Cost of bits and bytes is zero.

Speaker A: Zero exactly right. And there is no effort to make the bits and bytes once you've made it. To replicate the bits and bytes here. To replicate the bits and bytes, uh, the nuts and bolts, right. Is expensive. You need to put a person full time, right? You need to make an infrastructure around it. You need to shrink the time that it takes for him to produce that part. And he should be. Imagine if the car that you want to sit inside, how safe do you want it to be? Right? The weather conditions, the road conditions. A ton of material vibrating on top of that road that the wheel not running out on both sides, right? So it's not slipping on your rain and sun and so on and so forth, right? So you want that and for next five years at least they don't have issues with this. Right? That's the same expectation the customer also has. There is wear and tear. There is all of those things. What if the guy has made a mistake in the way he connects the wire middle gets damaged. M. You have that problem, the kind of pokey, okay that you have to think for each of those layers and structures where he will not make mistakes of what he's been. You can't put an engineer to assemble these products. You need anyone to be. You need to. And you need to think about automating through process first. Uh, that's another. Another insight that I gained which we have to think about is automation doesn't happen with missions. Automation happens with people on process first where you could put anybody given instruction set, anyone who can interpret the instruction set could actually repeat it. Right? That you're not dependent on one person if your technology demands you to be there to make it work. They have not built a productization yet. You have not built something usable in the first place. Right? So uh, factoring in where you can have people for manufacturing, um, where you don't need a highly skilled person to put your product together. So have you thought that in your uh, in your uh, in your, you know, the way you're putting, designing your hardware, how would you be able to mate them? How would you be able to mention them? You know, that's one of the first things that we even put as problem statements for people whom we recruit, right? Uh, we say is it manufacturable? So your design, functionally functional design is given for. Nobody's going to pay you if it's not even working first of all. But getting it working, just doing that particular task where it's able to pick and then drop here is not enough. That whole device. Have you designed in such a way that it's manufacturable? Have you designed them in the way that it is serviceable? Have you designed it in the way that it is scalable and modular so that something is primary and they keep missing this every time. And this is not a thought process that uh, comes right because it's not just about the manufacturability. Many times you get the piece machined in a certain way. You say that okay, this shape is possible but actually he will not be able to produce the precision on both sides. Like so. Simple thing is my camera needs uh, 20 micron of assembly tolerance. That's like 1/5 of a hair's thickness is the deviation that can have between two eyes. M. The more I increase this, the error that happens on this uh, between the camera in terms of angle and 81 times of processing keeps increasing. Mhm. Right now if I keep increasing this 81 times of processing every time, right? Say it does around 5 to 6 pixel deviation and it's multiplied by that. Now I need a processor, I need to achieve the task within the same time, within that one second, whatever it's supposed to do. Now I need a processor that is 81 times more powerful to achieve the same timing. Now if you put such a powerful processor in a factory floor, it is going to generate more heat. How do I remove that heat?

Speaker B: Right?

Speaker A: I need to have an active cooling, I need to put a fan and so on, so forth. Yeah, fan is miserable in uh, uh, an industrial environment because first of all it's greasy, it's dusty, it's oily and it's hot, right? So your fan will start accumulating all those gunk, gunk on top of it. It's going to slow down, it's going to start slowing Down. Then uh, the effect of heat removal is reduced. Then a process clock is downgraded, then the timing it is supposed to achieve is gone. Then the customer is not being able

Speaker B: to that five micron whatever.

Speaker A: Right. So this is one example goes all the way down there. Second, imagine the cost of never think that it's going to work absolutely fine. Your M product is going to fail for sure. In fact I say the statement saying, you know when your product uh, works only when your product fails.

Speaker B: Right. Because you figure the boundary conditions of

Speaker A: after uh, the product fails, is the customer going to come back to buy it from you or is going to just move on to somebody else or not use the product fail.

Speaker B: Is there a significant value that there is some tolerance to failure?

Speaker A: Failure. Right. Like if it is it failing on the same day that he bought, is it failing five years down the line and then after it fails is it coming only when he comes back? You know that you have your product market fit. M All this PMF statements uh, that they talk about and all that is you know in the air that you're borrowing some. When you translate that implementability sometimes easy to throw those terms.

Speaker B: Right.

Speaker A: Price market fit is very different from product market fit.

Speaker B: Interesting.

Speaker A: Right? Customer wants the product. Customer is going to benefit out of the product. But in what price? M He loves the form factor. It's going to work there. Tech market fit is very different from. So there are these layers that is missing.

Speaker B: Yeah. What I'm kind of imagining is everything is a component in this mega chain.

Speaker A: Yes.

Speaker B: Right. And if you increase costs, I mean uh, it is not just impacting your customer but your customers is customers.

Speaker A: Customers. And it's a chain. It's a chain. It's a chain.

Speaker B: Right.

Speaker A: It's loaded.

Speaker B: Yeah.

Speaker A: And depends on which layer in the industry are you hitting. Are you hitting just a step before the consumer or are you hitting at the bottom layer? And for a problem like this you don't know where all you would be hitting the tech like this where human hand is involved in all these layers. So everywhere there's a potential for your robot to come and play. So there comes that understanding. Uh, again. Right. So if it's going to fail, then it's going to cost every time it fails. Right. Can you put your service engineer every time sitting there and then fixing it all the time. And is it something that you can send a part to the customer and then replace it?

Speaker B: Mhm.

Speaker A: See, your customer does not care about whether the product works five years or not. Right. He cares about using Your solution without my production being interrupted. Can I continuously think on those lines? There is a possibility that you could put a service engineer there that's cheaper than you having uh. Do it charge on that. And if it's profitable for the customer on and above that. Great. Right. There are people who actually supply machines and along with that machine operator that they charge per piece this much PI say and so on and so forth. Even in India, like mpi, which is magnetic particle inspection, that they do for your. Your forged parts and casted parts. Whether they often to operate that machine, they also supply the machine supplier supplies a person, trained person along with it. Right. And the customer pays per piece cost to them. Right. In their facility. Right. Sometimes those are also possible. Those models are also possible. You can, you can. That's. That's left to your imagination.

Speaker B: But Right.

Speaker A: Factor this.

Speaker B: Right.

Speaker A: So.

Speaker B: So um. Again summarizing what you mentioned is that you have to think not just of the thing itself, but how it fits into this mega chain of components. Right. And you have to think in terms of systems and not just problem solution. Yeah. Right.

Speaker A: Ah.

Speaker B: Um.

Speaker A: And in fact one more point I missed. Uh, When I. When I was elucidating that you know the 20 microns. How much it is important.

Speaker B: Yeah.

Speaker A: Getting the 20 micron alignment. There is a certain ways of cutting that M part. Right. The material. There are so many ways to remove material. Right. Which is the way that you should adopt so that you can get the consistency of the tolerance on both sides. If that guy has to remove it and flip it and then put it 20 micron, how do you ensure that he is exactly in the same. He can place it.

Speaker B: Right.

Speaker A: Right. It's impossible.

Speaker B: Right.

Speaker A: Right. So then it's gone. So you need to have it as one single shot. Can you cut it Right. With the tolerances not changing so that then it changes the whole design for you.

Speaker B: Right.

Speaker A: And it. It affects your serviceability somewhere because you're not able to remove the part.

Speaker B: Mhm.

Speaker A: You can't just ship the part to the customer and you can't just take it and then fit it there. Right in the. Or does he have to ship the whole device here and you have to repair it here? Only here you could repair. I have to send it back. That's like couple of months gone. Meanwhile you need to send a spare and let's say your probability of your failure is very high. You will end up having a stock always for every system that is sold. M. Right. Your business is unviable again.

Speaker B: Right. How do you iterate in this kind of an environment. Right? Because it's not like your typical software fail fast, um, you know, ship, uh, um, just an okay, MVP of sorts and then build on top of it kind of a business. It has to work right the first time you do it and there's a significant cost to getting it wrong. So yeah. How do you iterate in this business?

Speaker A: Um, this is one thing that uh, you know when we begin it was the peak of all the sensationalization of how to run an organization fail fast, break things. Um, that is applicable Sometimes what happens is we take these thesis and then apply out of context. It's not universal enough. It's just these principles are within a particular context. Right. You can't have a heart surgeon and then say fail fast, break things or an orthopedician who is acting on your knee and then say that fail fast, break things. You need to be tarao before you could. You don't have the luxury of control Z.

Speaker B: Right.

Speaker A: Software enjoys the luxury of having an undo that immediately. So the cost of preparation is higher than just repeating. Right? Experimenting. On the other hand, the human nature of developing or designing something is incremental. We do something and then we start slowly. You can't planet everything, right. The way we have to think about is how do we augment with infrastructure. What are the failures that we have to avoid. You got to start thinking from oh it's going to fail, am I equipped to tackle it? If there's going to be a spark, do I have my fire extinguisher? Right beside if it's going to foul and then break the hardware? Because here if a hardware goes and touches and breaks it right there it's gone. Right. So your, your $100,000 just gone or $20,000 just gone robotic arm. You have to ship it back to get it all repaired. And repair is costly.

Speaker B: Right.

Speaker A: It might end up even harming a person. You can't press a control Z and then ensure that his finger comes back. Right. In fact this is the fallacy of uh, uh, accident right there. So this finger broke. It happens. This was uh, in a customer's place uh, when you were doing the three year thing. So it happens to ensure not to happen. Is first thing is do you prepare your environment and ensure that accidents are minimized and experimental outcomes are minimized? M. Right. How do you keep your spas if you don't plan for your failures? You need to plan your failures. You can't leave it to serendipity for a compiler to throw that error, and then automatically then, oh, I'll just react to that error, uh, debug and react to the error. You need to pre plan your failures. That's the only way by which you will. Right. So I know my robot is going to. There's a possibility that my code is going to, you know, make it go haywire. So what all do I have first that can prevent such a situation to happen? What all do I carry with me that I can immediately stop or I could? So you need to prepare yourself a lot more on one side, the second. This is industry where theory matters a lot. M. You need to look to that theory, uh, though they say, you know, practical, practical, you can't. It's. It's a balance between both. Right. Theory will not give you answers, will not solve it for you. Theory will ensure that your failures are minimized. Right. You need to be very strong with your discipline, go to the fundamentals much more, um, you know, easily, and then it should serve you when you want. Right. So that unfortunately there is a deviation of that thought process right now, these days, because you are able to arrive at solutions by just putting that up to an abstract layer, putting things together, and then you're able to solve things. So a lot of people's thinking is going into that easy, right? It's true in certain contexts, in certain industries, not so in hardware from that case. Right. So the only way which you can do is to plan your failures.

Speaker B: Right. And you have to solve it at the level of, uh, physics first and then engineering. Correct. Right. I mean, this feels like a lot to figure out, right? I mean, for someone. Is there some help that, uh, someone can get either through an incubation or, uh, through some kind of a government organization? Um, if someone's listening to it and just completely overwhelmed, what would you recommend to them? Yeah,

Speaker A: for us, nothing was useful. So I'm a bad person to. I'm not the right person to give an insight into it, but I've heard from the industry. So what others have, I don't have a firsthand experience of.

Speaker B: Because one thing being used I'm thinking of perhaps is that they go to a more mature organization with this idea, right? And say that, hey, this could potentially be valuable. I mean, can I use your setup? Right. And maybe you can take, I don't know, 4 or 5% of my equity or 8% of my equity. Um, right.

Speaker A: See, often you have to think about what, what do they gain by that? Right. And who's allocating that budget for them, how much of an experimental budget is that how much do they minimize those? So a proper access to hardware in itself becomes a lot of problem usually when you're sharing resource m. Right, Because I need to have my, let's say an oscilloscope. I don't know when my issue will be. I can't just rent for one hour and then think that within that period I'll be able to solve these things. I need to have it full time with me. I can't estimate, I can't estimate it, right? Because you are in an experimental phase. So thinking that and then the amount of bureaucracy that comes along with it often when you're renting, when you're somebody else is sparing something for you, you're responsible for it and they have systems to ensure that abuse of that infrastructure doesn't happen. So you will. It comes at the cost of that time being lost, right. Unless and until. So I'll break it down, right? So first thing first, of course this will be overwhelming, but there are simpler stages, right? Just for our sanity sake, we just break it down. To understand the industry, we just break it down into different layers, right? So application, platform, technology, you need to understand where are you playing? Are you inventing the tech in itself or are you inventing an application on the top? Right. So uh, some examples is uh, for example is let's say somebody like an Uber or ride sharing thing. Like I said, they are a business model genius who's leveraging technology then inventing them. They didn't invent the gps, they didn't invent the satellites, they didn't invent the roads, they didn't invent the tar. They didn't invent the car and cars, rubbers and engines and whatnot. Like the plethora of. And then ability to put that phone in everybody's pocket which is again somebody says business model innovation, Internet being available like JIO and then making it so cheap and then so on and so forth. So all that is a very different uh, leveraging of. So in those cases industry will already provide you with all this infrastructure. The know how will be there already. You don't have to source it. If m. You are in that category, these help might work out really well.

Speaker B: Right?

Speaker A: I'll skip the platform because platform is more of a hindsight realization.

Speaker B: No, as you mentioned that I mean I just, I'm thinking about how much we take for granted in the software world, right? Ah, we don't develop the security, we don't develop the infrastructure. We simply, I mean uh, we Just

Speaker A: innovated the application just for this building. Right. So you gave a compliment for this space. That how it has come out. I have to take this place. The amount of thinking that you do for real estate, right. There is talent on one side. The kind of talent that you have to attract here. Right. Customer base. And we anticipated like in 2021 when I was looking at it, I was anticipating it just post Covid because there is Mercedes Benz and uh, and jlr.

Speaker B: Ah.

Speaker A: And Applied Materials and Airbus. This is the R and D hub of hardware. Essentially Qualcomm. And so. And so. Right. This paid off now it's paying off this year.

Speaker B: M. Right.

Speaker A: That many of them are now coming in because the space is here and they're able to witness this. It actually works out. And for me to have this to convince, uh, uh, uh, you know, software, uh, Tech park to host hardware in here because they didn't even have a classification. Understanding that manufacturing is different. R D is different. M. Right. I have to pull out 6 inches of concrete entirely out and relay this whole space, the amount of investment that I have to make to just make this work. Right. Relay most of these things entirely. This whole place was not taken for more than four, uh, years before that. Some 2018 or so that they didn't uh, it was like a dump. Pick the dump, revamp this. Then showing us they just filled the rest of the other. Uh, so you have the COVID charm for uh, this uh, builder here. Right. You have to plan all these just to have this for this concrete thickness we don't have in commercially available power socket units on the floor. This has to be imported from Singapore again and then has to be brought here and then done and then later on. So everything. So it's designing every angle and aspect of. Plus it has to be a brandable space. When they come in, they need to have that of. And then translating, telling this to the builders. Eventually I had to sit and make the whole architecture, including all these, the colors, the material, the choosing of this carpet and then the static electricity that it will produce. And then how do you bring it in? I don't have a clean room. We don't have shoes here because uh, we can't afford a clean room. And the major source of dust is shoes. So we get them out and then plan certain layers of mats so that it sucks that. And then there is a cleaning procedure for it. So you have to map all of this crazy. Then if you. So, so yeah, so. So this becomes the, these, these inconveniences are not easily felt when you have that comes on the biggest lot of. I would say that the Dells and uh, the Intels of the world had been dumb. If you have to play for play for you know, App Store and Play Store because you're using them as a platform to build your apps, then shouldn't they be paying for Intel? And uh, the actual value of them is exploited through the software. And this is the amount of complexity that you'll remove when you do standardization of hardware. That's why the universal factory is a transformation. And what we think is it's not going to be mega or gigafactories in the future. It's going to be micro factories of universal nature that are sitting somewhere in your street end or under your apartment someday to both produce from the top and then uh, the bottom processing all the uh. So you'll be mining from the consumers rather than from your dustbins. The final consumer is your dustbin, not you even. Right. So you should be mining from there rather than mining from you know, the earth. Again why is it so. Because it's cheaper from ah, an effort basis when you make effort free or much cheaper than what a human effort is. You're not paying salaries to those missions and they don't have incremental bonuses that they need every year and so on and so forth. There's a lot of other possibilities that comes. Right. So I think many of these simplifications will occur in the material world if you standardize the hardware M and software enjoys the luxury of standardized hardware that somewhere else sitting and then coding it working exactly the same way on your system. And I'm not talking about the hardware as such. The platform standardization actually comes. So that is too early to do today on one side, because robotics is insufficient. They don't have every capability to say that it will cover all problems. So now we can sit and standardize it. Right. If you start standardizing now it'll become a bottleneck for technology growth. You can't too. So there is this chicken and egg the industry is going through right now. Right. So yeah, hardware has all these immense issues.

Speaker B: You mentioned application platform and then the third was. Third was tech. Right. Um, so where can people seek help at this point of time?

Speaker A: Tech, I think you're on your own mostly. Uh, that's how we were. You got to figure that out. You need really uh, you need an angel customer, you need an angel investor, uh, who's actually the angel who sees through this. And then we were lucky enough to have One um, you spoke to Arjun and yeah, so yeah they were, they were the. I think we thought that this wouldn't exist.

Speaker B: Right.

Speaker A: There's a lot of credit oaths vote to them and there's a lot of them, A lot of them helped. But it doesn't come into a proper structured material system. You have to figure that out. So when you are tech it's very hard when you're at an application layer where you could productize them, but it's an application layer of borrowed technology. These helps do come when you're a technology layer. It's a very hard problem. M and often that's why many of these technological innovations are. You wait for big organizations to come up with that way. We have pitched into a ah, space where it's uh, same problem Nvidia went through. One of the best examples uh, that I look forward to as a pattern is where most of the times we are contrasted against patterns that don't fit our thesis. And then um, they look at our organization as an underperforming organization. Right. But what they don't understand is building an industry like this, like for a gpu. Carving out a niche of an industry has taken these many years. It's a very, very hard uh, business, right? Very very hard uh, growth phase that they went through.

Speaker B: It's an overnight success that's taken about 10 years. Maybe.

Speaker A: Yeah, yeah, yeah. Maybe 20 years. Actually 20, 20, almost 25 years that they had to wait on this. So that's the patience game. Anything that is this deep, they didn't even have a market. How do you even anticipate market? Now comes the product market fit and all that, right? So you have to build the market along with it. When you are building an industry you have to build a market around it, right? And the use for it might have been there how it manifests. They needed a generation to grow along with GPU to start thinking applications for GPU which has become now a market for them.

Speaker B: Right.

Speaker A: They grew them from the college all the way till to become an industry senior and now to in the decision making position to think their problems through. GPU as a solution.

Speaker B: Right, right, right.

Speaker A: So yeah, you will have, you will face that. You know, brace yourself when you are thinking about technology problems that way.

Speaker B: M so let's talk about some of the technical challenges that you've solved. Right? And again, you know, for a lay person, lay person listening, like this whole um, implanting vision for robots, right? What are some two or three really um, specialized technical problems that you've Solved. Uh, that's been like your innovation. Yeah.

Speaker A: Um, I think uh, in the introduction I was talking about how foundational vision is uh, underestimated and under uh. Seen through your eyes do not capture the images like the way uh, your uh, your cameras do. Right. What you're looking at is color here. Right. What are you doing? It's all this world. You just put it, project it onto a flat surface. Colors that are there. You capture the colors and assume that I can figure out basis of color pattern. Right. Now if I ask you, this whole roof on the top, the side of the roof, it's all painted white. Right? You thought all of them is white. Right. If I take an image and then I look at it, that's gray, that's dark grey, that's almost black and this is white. 255. So like this. The color is a very subjective interpretation. The same object manifests itself. You rotate it as a different bunch of colors. Change the lighting in different bunch of colors based on lighting.

Speaker B: Yeah.

Speaker A: If I give you a coin, what reflections would you actually see?

Speaker B: Right.

Speaker A: If you have to read what's written there, you have to start tilting. Right. You will start adapting to it.

Speaker B: You don't have.

Speaker A: But how do you know how much to tilt? I don't know what to look at it. Right. Human eye. So the, the typical image processing world thinks from color. Then color patterns then match the color patterns to construct depth. Or M. Some patterns to construct depth. Human eye does not see color first. It sees motion first or changes first. Then it does not have just one way of constructing depth. What we say is if you have two eyes you can see depth. What is our foundational assumption? Right. We have monoclonal cues of depth. We have binocular cues of depth. If I get my numbers right, possibly around 18 different cues of depth is actually there. Six are very primal. Some are learned because you know the object. You are able to place them because you know that that's a corner of a room. Then you can actually place that as a depth. Those are all learnt cues. Very higher order. There are fundamental primitive layers of depth construction. There is close to around 6 you're converging of your eyes. Your auto focus is not just to keep things in focus but it's the first layer of depth that you construct. M. Right. So there goes these multiple layers of motion in itself. Like you've seen your cartoons, right? Uh, just simple 2D cartoons. But gives you an illusion of depth. This Doradora and all those kind of things. Uh, uh, where the Character moves, uh, faster. The trees behind them move slightly slower. The mountains move very slow. Three layers. The Walt Disney, uh, invention of how to create depth in a 2D things that motion gives you the depth. The reason why your, you know, your chickens and your pigeons do this, we think that they have a very low amount of, you know, convergent depth, which is the overlapping between your binocular depth, right? But they can do through motion M to a much larger extent, right? So, of course, there are other gates and things that actually comes. So many of these foundational capability itself is underestimated, right? So then we use the textures. When we use. How do we extract those textures? So you, uh, already start just like how depth is a constructed information, not a natural information that your camera already captures. You build depth, right? You process and build depth. The camera doesn't capture depth by itself. It's two individual 2D cams having no understanding of depth, but constructs them. There are several other layers of information that it actually constructs which you don't realize. It breaks down the thing that is there in front of it by how to understand this whole space from its surface reflection pattern and seeing the reflection pattern by which you define, stitch the surfaces together. If I put my hand below here, how do I know? Is it a black that is drawn on my hand, or is it, uh, the system has ways to, without knowing what that object is, to distinguish what's behind, what's in the front, where it starts, where it actually ends. None of these are learned cues. These are primal cues. If you have this, then you can say that an object that can distinguish itself from the hand which is on top of it is an independent object.

Speaker B: But if this is primal and it is intuitive, what is an algorithm that you can write to impute this intelligence to something like a device?

Speaker A: I think one of the first things that you have to go back is stop reading your AI papers, start reading your neuroscience. And neuroscience gives you the foundation of how your brain is actually working. If you want to imitate your brain, you want to start from imitating what are you trying. You don't have definition for what intelligence is. When the intelligence stops. And when intelligence starts, it's a very murky space. We are trying to imitate the behaviors of, or the capability for the human adaptation, human level of solving situations, right? Deciphering what's happening and then having a. Having a, you know, approach to solve it, right? So then you need to know how that system actually works first. M And there are so much more gaps, right? So often there's an argument. Are you saying, are you a nice sayer who is saying that that's not enough and so on and so forth. But my question is, are they not going to give you updates? M Are they not going to improve? Are you saying that everything to be built in the AI has already been built? If not, then what is the gap? Where are we actually having? Think from that outlook right now. If you see there are these. What I think is. See the first foundational assumption that your neurons actually take is in neural network. Neural network is legit. So it's a proper thing that you have to adopt. The problem is the simplification of neuron in itself. It's just a Y is equal to MX +C which is a straight line equation that you are putting it as wb, uh, wx plus B as bias and all that, right? So when you're tuning that, you're trying to take a complex behavior of a system breaking down into piecewise linear models. That's what you're doing. It's more of a mathematical approach than imitating what your neurons in itself are doing. And you're saying all neurons are same neurons, they're identical, just their weights are changing. That's not how your brains, between the first four layers of your eye, each of the neurons are entirely different. M from your V1 to V6, your LGN. All of these are very different functionalities. They already construct motion. Right. Just to throw some terms, um, basically they are called after your retina. There is something called as amacrine cells. There is something called as on of ganglions, which is at the bottom before that, on bipolar and off. Bipolar and horizontal. Sorry. First horizontal, then on bipolar, then amacrine. Then you have your ganglion cells on, off and all of them you can't change. Pick one neuron and then put it into the other place and then make it work. They're non identical. They're completely different. They have their own non linear behavior. They're not linear machines. If you don't just like that, if you don't have that structure, if you don't have those arrangements, what are the cues by which the system is supposed to learn? M right. How do I stitch the behaviors, eliminate the variations and then say the commonality of the same object in different situations. Then I say that, okay, I have to approach, go pick it like this, act on this, and so on, so forth, right? First thing you need dynamic vision. You need vision that can adjust, uh, itself and then collect more information. Then every time it fails Take the image, force it. Like beating a duck to lay golden egg. Right.

Speaker B: That's.

Speaker A: That's the approach that we're taking. Right. Oh, it fails. Now, if it's not. First problem is with the AI is if it's not able to detect that object, it cannot go act on it. M. Right. But a baby, without knowing what that object is classifying, without pencil or A or B, it can go pick. It can figure out how to pick it. Right. Which is not something that, uh, the system can't do. Can do. Now, if it is unknown, he can't show that. I don't know what this object is, but I know that this pickable contour. Can I go and pick? M. It's a pickable feature, the object. There is something that is pickable. I can pick when I pick. Now I know whether these are all attached or this or not. So definition of an object comes together.

Speaker B: Right? So there's some amount of processing, real time that we do. Right. And it's continuous rather than, uh, processing and then, like acting on that.

Speaker A: Correct. So there are these multiple layers, Right. For people who can understand, they will. They will notice these are issues that are actually there. Right. Several layers of assumptions on which you are operating. We are hoping that this will solve.

Speaker B: Right. So is it fair to say that you've built like a pair of eyes and a pair of movable, um, adaptive hands, and then.

Speaker A: So that's on the hardware side, right? What you do using that, how do you employ those faculties, those capabilities to arrive at all these cues that I'm talking about? M. So major work actually happens on that side of intelligence and the neurons that you're building and so on and so forth. So. So those are the layers. So in fact, we do not. We don't have the luxury of using what is readily available out there. We are forced to build most of our layers from scratch by ourselves. So that's another problem that we face. But it has largely has helped us to penetrate and solve problems that. It has been a huge issue thus far. Right. Given this basic layer of, uh, capability that we've built into the system, it's able to immediately kind of handle them.

Speaker B: So all of this complexity seems like it requires the brightest minds to be working on this and lot of interdisciplinary stuff as well. Right. I mean, there is, um, electronics, there's computer science. Heck, I mean, there is like, neuroscience and anatomy and God knows, I mean, maybe half a dozen things I'm missing. Right? Yeah. Um, how do you find this talent? Um, how do you nurture these, you know, these folks. Um, yeah, I mean how do you build a company around this?

Speaker A: Yeah, I think uh, there's a lot of leap of faith in this. Uh, I mean we can't expect the industry to supply talent for technology to be built. So that's, that's the first problem. So I have to divide them into the foundational principles. People who have to work from a fundamentals and fundamentals of the principles in itself. People who can build the tools and people who use the tools. Use the tools. You can actually. So this is again a sanity sake for us to say what we should look out in the market, what we should groom in house. Significantly, a lot of portion of the talent that you have, uh, the core talent is always groomed in us, be it on the business side. Uh, again, business fundamentals, what your MBAs teach and everything is, are they, they produce management folks who need something to manage in the first place. Right. So when you're building things, it's not yet existing. Right. And the processes and things are not universal things for you to the, the laws of processes that they are taught and case studies that they are taught are not universal enough to just borrow and be inspired about. Even often we imitate than being inspired.

Speaker B: Right.

Speaker A: So and when you imitate you interpret this different problem and then through the lens of the problem that you read before, which ends up in collapsing the whole company. So that's why I'm saying even business aspects we have to groom in uh, house the fundamentals of them. And people who come from another industry, they have been taught and groomed to think to ensure that the outcomes are very similar to what they want. Right. And those processes were built to ensure that it doesn't deviate from their outputs. Right. Now if he comes with that learning and starts applying here, I'll just become a umbrella company of them. I can't build my own industry altogether. So you have to invent. You need to think like a guy who invented the process from the scratch. So you need to start looking for them. So that's the, that's a foundation layer. The, the foundation layer, that of talent that actually comes everywhere where we, we break into what all the aspects that we need innovating, uh, or inventing from scratch. Then comes people who are really savvy, who could hustle, who could think in a very different way, but not out of a routine practice m like a C programmer, let's say somebody who wants to in India. The opportunity for them to explore to the depth of how your computer architecture Is and then optimize something and pull out the maximum optimization of it is very, very narrow. But there are people out of interest who would have developed the skill set for themselves. Right. So game engine development is one of those things which foundationally has that, uh, penchant to think.

Speaker B: So you have to find a proxy for what you are doing. Yeah.

Speaker A: What you're doing. And then you have to borrow from what is the mirror industry from which we can. Which has the same mirror problem. M sorry. There are similar problems there. The second thing is that fellow would have. But the bottleneck there is he would have thought of his career growth within that industry.

Speaker B: Uh, okay.

Speaker A: Right. It's one thing to have the talent available there. How do you get them aligned? Because I don't have a handle on them. If you don't have a handle on them, then it's not a tool. It's like a knife that cuts you more than it cuts the. You need a. You know, a knife. Sharp talent is a knife. Right. And it'll end up cutting you more double. It's what then. So you need a proper handle for them. So that handle is when are they motivated to come here and then do this? Is this industry resonating with them now there comes a huge portion of branding that has to happen in between which is again a pain. And uh, in spite of that inspiring harmony will not see this as a risk and move away from it. The convenience of how to navigate that industry is something they know. They have nothing that they can know here. And there is that fear to jump that to progress their career and so on, so forth. So those are challenges that will come along even to repurpose somebody from another industry into a new industry altogether. Together then comes the top layer where we are just using the infrastructure asset. We are also using a significant portion of it. We might be adding 10% innovation in terms of what's already available.

Speaker B: That's all.

Speaker A: Right. It's a huge change. Even the 10%. Right. So for all those 90% we will look for people from you know, uh, where. Which might be common. Right. Certain aspects of let's say hiring hr. Certain aspects of marketing for that case. Right. May not be at the branding but at the marketing level. Brand building versus the difference is at

Speaker B: the margin and not like fundamentally it's not the fundamentals. Right.

Speaker A: So and then we sometimes some roles are a composition of these. That's the most trickiest, hardest to solve. You know, the hardest, hardest role that we were. We were looking. We have been struggling to close Is it. Oh, IT manager.

Speaker B: Right. Or like a systems engineer.

Speaker A: Yes. That's the hardest for us to solve. Right. So then, uh, uh, I mean, who would have thought? Right.

Speaker B: So yeah, I mean, I would imagine. Right. I mean it doesn't look like they're just a bunch of pieces of max here.

Speaker A: People who are 30 years back will connect to this really well. Right, right. And the problem with India is most of those organizations which are looking at those IT personnel and things, they're borrowing, the processes were invented in their country of origin. Right. That's. Most of them are MNCs. M. And these guys have their whole career groomed within somebody telling you what all the processes were.

Speaker B: Got it.

Speaker A: If you have to come here and innovate and then build a completely different set of trust. So we keep saying we need an information manager, not an it. He looks at all the information. Basically everything is about information. It's flowing from one department to another department. What all tools that are needed. How do you integrate the flow? How do you make optimizations into that? A person who can come and look at guy, uh, scrolling at 10 times. How do I reduce that scrolling? Because each one taking 10 seconds is. If he does it thousand times in a day, it's 10,000 seconds gone. Right, Right. So that's, that's a lot. That's a lot of productivity loss. Is someone thinking. No knows to look at it. Think that. And then bring processes that could. And tools and infrastructure that could simplify the problem.

Speaker B: Right.

Speaker A: So all this is a. Is a.

Speaker B: Right. A lot of first principles thinking.

Speaker A: A lot of first principles thinking.

Speaker B: Right. Yeah. How do you approach fundraising? I mean, um, Special has invested um, in Signler. Um. Right. So any advice on how to fundraise for uh, like a deep tech slash manufacturing, slash robotic kind of a startup?

Speaker A: Again, I'm not the great guy to uh. No, whatever you advise on this. But yeah. What has happened through us, there's a lot of serendipity that is involved into it. Even the current round of lead investors, uh, which is Pavestone and Athera for us. Right. So all of them come from a different background. There are sometimes the market alignment that happens, uh, where if the humanoid sensation and deep tech sensation. I don't think it's about the sensation of the deep tech. Many of these investors are uh, thinking from the fundamentals. Some of the investors who are thinking from fundamentals who are not riding the trend. Trend does influence, but not like they're reacting to the trend. Right. Do ask the question of what else is there now to invest, right? So if you look at the market 90s are all about building the funnels for Internet. And then the late 90s and early 2000s were exploiting Internet. Then.com bubble came and then it fell. Then after that was again mobile device revolution and communication revolution was the biggest. Then digital computers finally digital computers nailed it was there in everything, right? Uh arm processes and things. That was 10 good 10 years. Then came another market that is exploiting that infrastructure that's been built, right? That's when the companies like Nvidia and all of them were born during that phase. And then this 2011 till 2021 was all about exploiting this. And Covid tested the peak of what you could do with this. And then it just came into standstill. And at the time the trend was your zooms and uh, what could enable remote work was the biggest uh trend. Uh in fact in 201718 there was one lullaby we thought now the industry because the app and app uh, aggregator and marketplace aggregation started taking a dip. E commerce and all of that started taking a dip because everything to be done there are already players who have already captured it. There's nothing much to capture. We thought now they will look at deep tech. And that's when we started looking for funding in 2019. But SAS came up right? They started looking at okay this infrastructure is done but it has created a deep uh, uh industry under it which needs a uh support ecosystem further than the SaaS. Software services started picking up so much more. Then came these work from home enabling systems and things. At the time this was a conversation that used to happen within the VCs an argument that just happened that I was saying supply chain is going to come out being the biggest bottleneck because the fundamentals are still there. M the market takes some time to react to it. But you see in 2021, 22 around that time the technology needed to support supply chain will come and that will be more deeper problems. Lucky enough there is this along with open air, there is humanoid thing that is trend that is catching the issue with the trends is when they fail uh, it collapses the market along with it, right? So you can't time it also and it's not in your control, right? So this is where it's a risk to invest in an organization like ours, right? Second, it's not just the VC's intention, it's also where they are raising fund from and what is the patience that the fund allows them to do. Where do the LPs want to kind of focus on this and what do they say as thesis onto which they want to invest? So if there is not an undercurrent of a narrative that's moving, you need more patient capital. You need more patient capital. Right. Third, the markers for them to judge you M. Because it's a uh, it's a. See either you need the investors who just think that okay, I'll drop it, I'll do an experiment. If it fails, I don't give a damn.

Speaker B: Right.

Speaker A: That's a risk for you because they would have just got you up to the next stage. Possibly the current money already would have got you up. Will they back you up after that? This is an indicator for you to raise next, right? Do they are how invested are they into this industry that actually matters now if you put all these filters, there is very, very niche set of investors that will be there. M. Right. And that money doesn't have category into which it has to invest. Right. Anybody can compete for that money. Any business can compete for the money. So that's a tough game. Last time, um, um, I mean this race mostly nickel focused entirely on it. It was nickel's thing. And then even uh, the RR pre Series A that we raised uh the tune of 4.25 million which again reduced now because of $inr uh ratio. But um, uh, we both did close to around. We interfaced close to around 114 investors or so pitches would have been like around 150 or so pitches that you would have odd which is that you would have done each one going above an hour. Imagine the whole year just goes wasting in the time.

Speaker B: Insane.

Speaker A: Right. And that's how we split saying that okay, Nikhil is going to focus entirely onto this and then I let me focus on building. His whole time is completely gone into this.

Speaker B: Yeah, no, fundraising is a, is a full time thing for sure.

Speaker A: And knowing those investors second, you being in this geography for us is not helping so much. And then we having to go to us and establish the trust into this industry. Right. In us it has a very bad robotics has a very bad reputation. It has not been a very successful investment thus far. Right. Again you had Boston Dynamics, Covariance, vicarious of the world and everything hasn't paid to the hype that they anticipated with. And uh, this actually brings to the problem of looking to invest in the solutions rather than looking to invest in the components.

Speaker B: Right.

Speaker A: Like I said, the robots need so much more sensors. The grippers are so poor. The components needed for the grippers are so poor. The Motors in itself has to be renovated. Right. The material needed for you to make the robotic arm has to be renovated.

Speaker B: But you would think, right, that let's say a Google with as many billions of dollars on its books would just like throw, you know, throw a couple of billion dollars here or there.

Speaker A: They also have to throw for so many other things that they have m. Right. We are just looking at one small narrow portion of their business. Right. The gamut of businesses that they actually have. There are so many different verticals and divisions. The Internet infrastructure in itself has to take up so much. So much of the software fundamentals and tools that they have to build in itself will take a lot more. There is a market expansion that will have to take a lot more. New market creation, which is outside of their strength is so much experimental that it will be very narrow. When you are having that narrow choice of money, you can't do wide variety of uh, trials. So then even those guys, right. So one of the, one of the largest e commerce organizations that we were looking at, uh, we were looking at us from a funding point of view and one of the statement that came from us, these guys are taking an approach that is entirely deviant from how it is. And I have already a 200 member team put into this.

Speaker B: Right, right.

Speaker A: And then they have worked for four years. Why would they want to throw that away and then pick yours? So there are all of these complexities that come along in various investments. Investment is not a technology decision. That's a business decision at the end of the day.

Speaker B: Right.

Speaker A: And they have to see what is the amount of investment that has already happened into it. What's the precedence for it. And I think don't look for the history of investors who have a tendency to keep investing in these spaces because they can do only one. They're already done. They're not your uh, investors. The ones who are yet to make that mistake are the ones who are. Whom you have to kind of go capture. Right. So I think and it's that change of mind where and you will be, you'll be undercurrent of an investment across a lot of these things. Expect that if it happens to be, you know that there is focused fund for you. That's great, Right? Right. He has to make one bet and he has limitations for only one bet.

Speaker B: Very interesting, right. I mean because like the typical VC mindset is to like fund, let's um, say via a thesis. Then you fund many different marketplaces for example. Right. Or many different SaaS companies. And so on but here you would be the exception to like the portfolio.

Speaker A: Yeah.

Speaker B: Than uh, the, than the, the entire fund thesis itself. Right.

Speaker A: So you need to look for the capital that has been allocated for the exceptions.

Speaker B: Yeah.

Speaker A: And somebody who has a very strong thesis most probably will not turn out to be. Because they will stick to their thesis and if it has worked out.

Speaker B: Yeah.

Speaker A: They will stay with the thesis unless and until that has been exhausted. Now um, if there is a moment where that exhaustion is happening, most probably then people are looking and then interfacing and then learning. 90% of your time will go in this pitching and then educating them.

Speaker B: Right, right.

Speaker A: And in fact they used to say right, if you have to educate the customer then this is not the right market. Right. Timing for you.

Speaker B: Right.

Speaker A: I have to do that only for my uh, uh, investor vendors or customers. It hasn't been that hard. But having the fuel which is the currency and the cash. I mean investors at the end of the day are vendors. Right. I have 400 parts as supply chain into my hardware. 401th part is if uh, you have a whole car, you need fuel.

Speaker B: Right?

Speaker A: Right. So that's another part that you spend on. Right. There is a cost to the money and then you are, you are getting that money.

Speaker B: Right.

Speaker A: Essentially as a fuel. But you need to look at somebody who, who, who is not, who's not stuck to the thesis and see who exception capital. Yeah.

Speaker B: I mean I think we have covered a lot of ground and a lot of the things that you brought up. I mean I definitely have a you know, follow up questions on that. But you know, we have our limitations as well and, and have taken more than um. Enough of your time on a Saturday. Um, you know my final question, right. And it's pretty random. You take it whichever direction you want to go. What do you see as the future of humans?

Speaker A: Oh, okay. That's a. I, I don't know uh, frankly. But see I don't have a dystopian view towards technology and I'm sick and tired of all those movies that gives you uh.

Speaker B: Like black mirror types.

Speaker A: Yeah, black. I don't have that. Yeah, right. So likewise, it will be a, It'll be a. I'm not saying there will not be any people will. Every fire is a double edged sword. Right. Everything if you don't use it. Right. It can go either which way. So that's the thing. I what I think I can say what could society be then how. What could be the future of humans? Humans will be there because these machines when you're working inside. You just realize that these are just dumb missions. It's just a tool. When people used to think that when cars came in 1920s the same fear was that oh, humans will all be, you know, replaced tomorrow. Everything is going to be automated and so, and so it still is going on every mission. When they, when it came and simplified human effort and removed the necessity of a human in certain places, it's a, it's a projection. That fear is a good thing for the society sometimes, right? But sometimes it's over exaggerated. So I don't think something is going to happen on that line. We are going to learn to use these tools. These are more sophisticated tools that you are getting. The purpose is still ours. They don't have an intention to survive. We have uh, that's a foundation instinct, right? Even if you program them to be, they will still be dumb, right? They are not uh, I don't know when it will happen to the level where the mind is and the instincts are being programmed into the system and so and so forth. But uh, society will have significant transformation in, in, in the way like when I was saying we, this is all you. If you see these posters, this is all. This is the science fiction that we. This is, this is the vision map. I don't believe in like vision mission statement. I just think that it's a story and there is a, where people can imagine themselves in the science fiction and they have to work towards that. Science fiction, right? So when we think about how um, a transformation might happen, uh, in the way like the, the way the technology today has transformed, uh, I think there will be more of an assetless society that will come into place. Humans will be spending more on IP value creation than on physical effort being valued more. Uh, physical. I'm not saying physical effort will not be there, but physical effort will start getting minimized, monotonous, repetitive physical effort will start getting very, very low. And um, what your, your value of your brain and thinking will be much higher. And I think a lot more population has to be relieved to be able to think, sit and think and create more.

Speaker B: Right?

Speaker A: Um, on the other hand, when we think about this object, computers, right? So when you said when data computers came m, you had an Infosys kind of an IT company coming up and huge employment opportunity and so on. So, so what would that look like tomorrow, right? Somebody sitting imagine instead of laptops, this whole tech park is filled with these kind of robots where you're teaching robots which are programs that have been created for somewhere in Some end of a factory that they could download and then start using. It could be even consumer level that imagine have this uh, at a bigger picture you have this robot in your home and uh, the recipe that you're eating, you know, you went to Thailand, you loved the recipe. Today if you have to come back and eat that, that guy has to come back here into India, set up a brick and mortar setup, figure out a whole supply chain of material vegetables, uh, that this and masala and all that he has to bring in here. He has to train cooks, he has to operate it here, get the capital for all this, have at least 100 other customers every day so that they can serve this one customer interest.

Speaker B: Right.

Speaker A: On the other hand, if you could distribute all this that through an optic fiber cable, that recipe comes all the way here, right. And then your robot instantaneously uses feed into your context. Just picks a things and then makes it just pay for that one dish, right. Suddenly the recipe that you invented has a global reach the market. And then if you can do this in your robot, you could also take this robot out of your house. Your houses may actually lose the kitchens in the future. Today you may m not like swiggy food and all the day you can't eat it because it's a centralized industrialized kitchen and they are optimizing recipe for business model. If you could remove that cost from um then simplify this. You could you know, immortalize your, your grandmother's recipe through those rewards and then they could just keep making them. So. So from there to thinking of your how your house structure would be, how you are dispensing of your garbage would be and how the. How those setups will. That's. Those are the imaginations that it goes. It's a far fetched imagination. I don't think I'll be alive to see many of those. But the seed of that is how the industry is the direction the industry is going today. Right. And that's what I think and uh, the relief of the space and how we will go into efficiency model than exploitation model that today to sustain population we have to keep converting existing new material. We need to start gulping more and more of material from the earth. Uh because effort is cheaper there. If you can remove this arbitrage of effort cost to be simplified and then energy becomes a major cost. And the cost of making renewable energy also is very costly because the labor cost is still high. To make that all that. If you simplify the economy, the value. The value economy being circular starts the Consumption is only through energy that comes in rather than material. Today it's a lot more material consumption oriented economy. Right. That material will be a circular system. Energy is what you.

Speaker B: So energy will become like a currency, basically.

Speaker A: Energy will that. I don't know. Mind keeps going there, but I don't know, I have not thought through it that much. But maybe energy could be the currency. Maybe.

Speaker B: Right. All right. I think I've lost half my voice even though I never spoke. But, uh, yeah, on that very optimistic note, we come to the end of this podcast. Thank you so much again for spending your Saturday with us. And thank you team as well for, you know, their patience. M. This was a fascinating conversation. Uh, some of the best podcasts are those that like, you know, create all these rabbit holes that you could dive deep into. And certainly I'll be reflecting on this conversation and going into those rabbit holes. So thank you so much.

Speaker A: Thanks so much for those, uh, questions that triggers the thought processes and the way you set this up into kind of bringing the thoughts rather than not a template set of questions. Right. So at least that brings out the thoughts. So, yeah, thanks. I enjoyed this conversation.

Speaker B: Awesome. Likewise.

Speaker A: Thanks.

Speaker B: All right, folks, thank you so much for joining us. Uh, we'll be back with another episode of the Startup Operator podcast soon.