India GameChanger recorded and interesting conversation with Vikram Gupta and Yatin Kavishwar, Co-founders of Awiros. Awiros is an open platform for computer vision and AI developers, providing them with all of the necessary resources for transforming AI algorithms into end-to-end solutions.
Some of the topics Vikram and Yatin covered:
- What Vikram and Yatin believe will happen in two years
- People who are really serious about software should make their own hardware – Alan Kay
- The importance of having complicated skills
- How having the right connections can help open the right doors
- The impact of celebrating small achievements
- Vikram and Yatin’s unique toolchain
Some other titles we considered for this episode:
- Being Market-Ready and Adopted at Scale
- Start Behaving Like an Appliance Rather Than a General Purpose Computer
- Benchmarking and Building a Right Mix
This episode was produced by Stephanie Ng.
Read the best-effort transcript below (This technology is still not as good as they say it is…):
Michael Waitze 0:03
Hi, this is Michael Waitze. And welcome back to India GameChanger today we are joined by Vikram Gupta and Yatin Kavishwar, who are Co-founders of Awiros. And it’s great to have you on the show. How are you both doing today?
Vikram Gupta and Yatin Kavishwar 0:16
You’re doing very well. Thank you. Thank you might feel it’s a great opportunity for us to be here as well.
Thanks, Michael. Thanks for hosting us. Awesome. It’s good to be part of your podcast.
Michael Waitze 0:26
Thank you, it’s great to have you both. Before we get into the central part of this conversation, I like to get a little bit of the background of my guests to start from Can we start with you,
Vikram Gupta and Yatin Kavishwar 0:35
my story is actually quite simplistic. Our house is my first and only job. I’m an engineer by background, I did my PhD from Carnegie Mellon, which I finished in end of 2014. The goal was to build a technology platform to build the software infrastructure of the future. And I finished that PhD in kind of a concept of that, wanting to take that to market. And that is how we Ross was born and has an OS for Internet of Things. I can look into that I knew immediately I want to do that. And it came back to India and founded Eros. And that’s all what I’m about. So
Michael Waitze 1:11
all through my it’s so specific. It’s so specific. Yeah, Tina, how about you?
Vikram Gupta and Yatin Kavishwar 1:15
So I think I come from a completely different background. Vikram is more of a researcher and I spent 17 years in business development, sales, alliances, marketing, you know, before me and Vikram met. So I started my professional journey in 2000. So I was fortunate to work with companies like IBM, CA computer associates, then Cisco, Citrix, and my last experience was with AppDynamics. That was in 2015. In fact, the same year where victims started his journey, paddle me. And after 15, you know, I moved out and evangelise on, you know, technology and obviously, you know, with the intent to start something on my own right. However, when I saw the landscape being into it, so you know, I thought I need a technology partner with me. That will be very important aspect of our journey. That’s how I mean reconnect. So that’s interesting story. But I think this is good to introduce me at this point in time.
Michael Waitze 2:20
Can I ask you this? What was it like going to school at Carnegie Mellon in western Pennsylvania? And like, how did it make you look at the world in a way that was different? If maybe you had just studied in India for your entire life?
Vikram Gupta and Yatin Kavishwar 2:33
I think we I went there as a dreamy eyed boy, right? We I didn’t know what to do. And it was actually my first time out of India as well. Oh, wow. never travelled outside of India. I never knew what the Western world I think from India perspective looked like confer that I had I already had a few friends going to school there at Carnegie Mellon. So this was there was some microcosm of of support already available. Right? When I went to Carnegie Mellon, I realised that first thing that hit me was was the amount of pride that is involved in a western education, especially in the tier one education like CMU Yeah, it was amazing. So for us, to be honest, right? So we India, engineering was known to be okay, it’s a bunch of smart guys, but they will probably be relaxed about education a little bit, it will not be that much of involved. per row, I would say, learning really gardening it will and change that completely. So first semester was a big learning curve for me, I was not ready, I actually could not anticipate the environment that was going to come on towards me. And with PhD, it was two things that had to do in parallel, right, we had to do a few courses as well, as well as the start working on my research. I was fortunate with my with my peer group, they helped me out there, but it was a shock in the way even normally, computer science courses were done. And it was a big jump. I am sure. Fortunately, I failed.
Michael Waitze 3:59
Fortunately, I made it. I want to I want to ask you guys this too. I was doing a recording yesterday. And this just popped into my head. Yeah, I was doing a recording yesterday with a guy who’s I’m trying to remember this was LinkedIn profile said The future belongs to those who see possibilities before they become obvious, right? And it reminded me because I remember this from yesterday’s conversation that I said, one of my best investor, clients once said to me, the best investors are those people that anticipate other people’s anticipations. And I’m really curious, right, because you started it was, what, seven years ago? 2015. Can you both maybe remember, what was that thing that you saw? That wasn’t that obvious to everybody else that said, We’ve got to build what seems to me and again, tell me where I’m wrong. This integrated hardware and software company, right, so building an OS, tuning it specifically for specific hardware that you helped design in partnership, right. Like What Did you see that nobody else saw back
Vikram Gupta and Yatin Kavishwar 4:55
then what I saw or what I thought was, we knew the kind of financing Usually my PhD thesis was for such technologies to become market ready and being adopted at scale, right? software infrastructure has to come into place, which allows customers to leverage underlying infrastructure in a more efficient way. So I think I am not the first one to really invent this idea. But it has already happened in few areas before, right. So mobile phone and smartphone, that germy was a good example for us to take forward from a normal mobile phone in early 2000. Was just to make phone calls or maybe to text text messages, right? Yeah. But then smartphone came in marketplace of apps came in, integrated hardware came in, which lets you do so many more things that you couldn’t have even imagined. And we thought the same thing is supposed to happen in how we automate the physical world. And we believe this should happen. The technology was also changing towards that. But this thought that technology should enable large scale adoption of automation, especially through video. Yeah. And scalable in a scalable way, is something that is missing in the world today. And we believe, actually, to be very honest, we believe it will happen in two years. Yes, we were a little too optimistic, I guess. But that’s right. It did. So that inflection point we saw did come in in end of 2019, early 2020. And I think Japan has a very interesting perspective on how Hardware and Technology also played in. Go ahead.
Michael Waitze 6:31
And then I have a follow up.
Vikram Gupta and Yatin Kavishwar 6:33
Yeah, sure. So I’m not a techie right, or an engineer. But I think I filled that gap with my view towards how technology at large is adopted, right? That I was fortunate to have that experience with the companies that I worked with, I think during the same time to consulting 16, I got introduced to video AI, right as a concept. And parallely, as Vikram was building up his vision over his PhD thesis, suddenly, you know, when I got introduced to this technology, the interesting thing happened was visualisation, that can video AI be also adopted at scale, number one, and it has to be adopted at scale. What are the ingredients which are required? Now, obviously, as you rightly said, integration of the software with the right hardware and building those right tool sets, so that it can be scaled for adoption was actually missing. And in that endeavour, I came across another startup, which I started consulting, right. And, in fact, we’re trying to build up something around the same on the same lines, I think analogy, you know, what struck me during that time was the Android or the way Android and iOS have transformed adoption of applications on a mobile phone. And a similar platform might exist somewhere, or can it be build, which can help customers scale? The video AI applications? One more interesting. Timeline was again, 2015 16, where computer vision which is the core technology on Eros works, right. And artificial intelligence, they started getting commercialised together, right, that helped customers to build a lot of predictability in the application. And the adoption was obvious after that, right. So why I said, you know, brought the stories. Interestingly, when me and Rekha met, and we had our first conversation. And when I saw what Vikram is really doing, I think, all that my, you know, zeal and enthusiasm towards the video AI and the way I thought I think it matched picture perfect. Yeah. So I think that was one point where we both thought that yes, I think if there are two people, both from different diverse background, when getting the technology aspect, I’m more coming from the market. So there is a market, which I believe become said yes, there is a market if this product is there. And I think that was interesting, you know, handshake, you know,
Michael Waitze 9:15
we make some analogies here just to make sure that everybody understands this. First of all, it feels to me like there’s a through line with what Silicon Graphics did back in like the late 80s and early 90s, right, we want to build a machine that optimises for this one, one function, or highly optimised for this function. And we want to build a machine just for that. Now, they did not have access to the same throughput connectivity and compute that we have today. So it’s it’s a different thing. But if you look at what I call functional abstraction, right, let’s take something that people use technology to do, but they use it on a general purpose machine. Let’s take that function that we think is going to become way more important. Let’s abstract it away. And let’s build the hardware and the operating system. This is a really nuanced decision to make not just soft Where, but to build the OS for it means you’re making a bigger decision. Right. And in a way, what you’re saying is, we know we can build a bunch of software on top of this platform ourselves, but the platform business is always going to win. Because what it does is it democratises, the ability for other people to build on top of the platform, and then the platform never goes away. And even if we can’t build every piece of it, we’ve learned through the development of literally windows, the tight application of, of Apple of Mac OS, and then into the smartphones, that if you own the platform, you own the business. Is that what’s happening here in the video a, I suppose.
Vikram Gupta and Yatin Kavishwar 10:39
I think that is exactly the journey and the goal that we are following. So we believe that for this technology to get to its true potential, the underlying challenges. So there are many every technology product has vertical technology stack. And the core value add that the technology brings is actually in probably somewhere in the middle in the top two, three less on the right today, just hypothetically, we are using zoom for this call, it runs on so much of existing tech, right iOS, or Windows and the network stack, and so on the hardware that we’re using it for and so so it would have been impossible for zoom, or any video conferencing software to actually build it from scratch and take it to market. Right. So you cannot build a laptop for zoom, it will not fly. This is exactly what is happening. Right? This is exactly what has been happening in video AI or the space, we took a conscious decision that okay, we are going to do the heavy lifting first for us. And then for the world, that the technology stack that involves probably 10, large, complicated technologies. And I think in total about 120 and 30 moving parts, and we will give you five or six levers just to play with so that you can take your application, plug it in, and run it at scale without having to kind of reinvent the wheel every single time, like every other software abstraction, or platform has attempted to do in this space. We believe we are among the first to do and we want to take this to a global scale from here on
Michael Waitze 12:16
what are some of the difficulties of building your own? I mean, software itself is hard enough to develop, right? I mean, it’s just super hard to do. And it’s super hard to do well. But now you’ve added the complexity, and the sophistication of building hardware as well. Like where, you know, I’m trying to remember the famous Steve Jobs quote, but somebody said, if you really want to be serious about software, you have to build your own hardware. Right? So I completely understand this concept. But it adds it does add another level of complexity, right? I mean, it took decades for Microsoft to go, Okay, we’ll build the surface, right? And it was just after years of watching Apple eat their lunch and some of those devices. I’m not making a political decision here. But you guys said right from the beginning. If to do this, right, we need our own hardware, what level of complexity does it add to building this whole thing? Because the what’s the right word? The CapEx for building hardware is super different than the capex for building software. Yeah,
Vikram Gupta and Yatin Kavishwar 13:06
absolutely. Luckily, we were fortunate to basically with the Athens inputs. So we were fortunate to do hardware without a heavy capex. And this is also something that we were able to do well. So one of the things that we actually did was that we focused very deeply on partnering with the hardware, OEM. So we went to Intel, we went to Nvidia, we went to Cisco, we went to Dell, saying that, hey, this is our vision, we want to build this platform, which should work tightly with the underlying hardware without so that the customer can adopt it in a plug and play manner. And, and of course, we did, actually, we didn’t have any success story. But our vision and concept did resonate with some people. And we started benchmarking and building a right mix of I would say, hardware platforms that could become the, I would say, start behaving like an appliance rather than a general purpose computer.
Michael Waitze 14:07
This is the whole point, right? So if you if you own the hardware and the software at the end of the day, it should be way easier to use, because you control the entire experience. Is that
Vikram Gupta and Yatin Kavishwar 14:15
exactly we could Yes, absolutely. So we as I mentioned, we were fortunate to work with, with the Giants in this space, even a small startup and they are a multi billion dollar company. So we got the right support from those companies to build, not build, I would say to package the right specifications of hardware in a way that makes it easy for us to put our hardware on. So there is actually we have a few white papers in this space with all of those companies and, and we have kind of so that helped us do a hardware and software infra without the capex challenge. I think this is also something that we are probably, I would say among the early adopters of this approach in this space yet and I think you need to
add to this I think we were right, you know, at the right place at the right time with these, you know, companies, in fact, 2018 19, obviously, you know, when, when companies like Intel, Nvidia, as I mentioned, obviously, Cisco and Dell as well, because they are the guys who package everything in hardware. They were also, you know, quite eager to look at new use cases. And I think video AI, obviously, you know, video AI requires a lot of hardware, you know, to run in from an inferencing perspective. So, you know, for us to partner with them, was there a need for them to partner with us? Answer is yes, the company to go behind for these big guys, you know, company into donating, we were like 11, or 12 people company. So, I think we were really quite excited where the same was reciprocate from the other end. And I think that’s where we accelerated our journey of benchmarking and moving towards standardisation, because most of the enterprise customers, if they are looking at adoption of this technology, I think they would put a bet on companies which have done some work, some benchmarking, and I think partnering with them, really helped us, you know, take that whole position in the market, King walked
Michael Waitze 16:14
me through a little bit this process, this is something that’s super curious to me. And both of you said this, you know, we were just like an 11 person company, dealing with giants, you just don’t wake up one day and go, Okay, I’m just gonna call Intel and just do a deal with them. Like, that’s just not the way it works, right? And then do the same thing with absolute Cisco and video, you just don’t die like one 800 Deal Guy and just get that guy on the phone and do the thing. I’m really curious what the process is like, and I mean this, right, because I don’t know about finding the right person to talk to maybe getting a little bit of a run around internally, like, what does that whole process like? And how long did it take, right? Because for a startup every day is another dollar potentially spent and invested every day matters. So I’m just curious what that process is like.
Vikram Gupta and Yatin Kavishwar 17:00
Obviously, I think, as I said, you know, me, and Vikram, you know, we have lots of complementing skills, right? Become obviously, you know, he’s doing the heavy lifting on the tech. However, you know, when I decided to take that leap and jump in this decade domain, I already spent 17 years in the industry. So stay connected. Apps, yeah, that my network helped me at least, you know, open the doors and or not open, in fact, knock the right doors to begin with. However, when we you know what, and it was absolutely sales pitch that we made to these companies, when they saw merit, I think that’s where things started changing. However, I would, I mean, we were fortunate enough to have that network in place, so that we can cut down on that legwork and spending time to find out the right resources or the Connect. So yeah, so I think this is one very, very important aspect. And obviously, once we had the right connect, established, I think our tech team did a phenomenal job in terms of turnaround. Yeah, the expectation was, obviously, you know, we get only one chance, and there are many in the queue. Right. So that was the only window, what what helped, was to turn around immediately, and show the real results and outputs. And obviously, you know, that is reflecting in the white papers that are published, not on our websites. But all these four companies, you know, we have likely for at least published now,
Michael Waitze 18:24
currently got it the premise there kind of and I say this along with you, right, but is there a kind of geek joy in this optimization process? Do you know what I mean? Being able to take software team is written in the hardware that the team has built it just make it better and better and better. So then when you go to Intel and Cisco and Nvidia Intel, right? They’re like, Wow, I can’t believe how much this has changed. Is there some kind of geek joy in this? You know what I mean?
Vikram Gupta and Yatin Kavishwar 18:48
Absolutely. So actually, this is the only joy that has actually kept us going to be honest. Right? So you cannot imagine there are some of the things that we discover still, that there are these tiny windows of optimization. So I’ll give you one example. very recent phenomenon, right. So we will now testing our platform on running at a large cluster. So instead of running on one device, now we are running it on, let’s say 10 large Dell servers together, and we want to monitor throughput of every single component in the system. So the Intel CPUs the NVIDIA GPUs, and one fine day we were actually looking at me and you think we’re looking at a screen and We’re suing this one piece of hardware is not giving the right throughput. Why? Where is this bottleneck? So we were actually brainstorming, we thought maybe the network card is getting the bottleneck here. Maybe the motherboard is a bottleneck, maybe it’s not the right configuration. We were we were like shellshock because this is not this was not something that we expected. Right? This is not I’m not talking about four years ago. I’m talking about two weeks ago. Earlier, we did that, but we never went into that detail that what impact I’ll tell you What happened was that so there are buses on the motherboard like PCIe buses, right. So their capacity actually matters when we are running things at the scale. And we were shocked to find that except for probably very small community somewhere in hidden in the in this internet world, nobody had actually anticipated this problem. So So one GPU or one Nvidia device is only getting a certain limited bandwidth from the motherboard and the CPU as the entire hardware chain, the PCI bus on the card, the PCI bus on the motherboard, the PCI channel supported by the CPU, all of them have to match and kind of be able to take up that workload simultaneously. If there is there is a delta anywhere in this middle, right? If there is any one weak link in the chain, the entire chain actually slows down. And that is what was happening. And we were trying to dig out for them and where it could have been happened. It could have been done at scale. We couldn’t we couldn’t find any. So this was a discovery. I wouldn’t say it was an invincible discovery, which which gave us a lot of joy. And I was actually running around the, with the team that hey, this is something that we have found and I went to Japan because actually coming from it isn’t see what you found. Why the behaviour that we’re seeing? Why was that happening? We were able to kind of figure that out. Yeah, so that child is joy of figuring out things is a primary mover. For us. It sure
Michael Waitze 21:27
is. It’s a you can hear it in your sorry. Yeah.
Vikram Gupta and Yatin Kavishwar 21:31
Mike? No, I was I was trying to add, you know, some sales joy, also. So, yeah, so, you know, we were just a team of 11 1011 people, and every single small achievement, you know, was celebrated. Right? I remember one incidents, you know, this was a third meeting with one of my sales team members, and me and him were, you know, presenting to third customer? Why is it third, you’ll understand the first meeting was when we showcased our Cisco white paper first time to one of the customers. So I went to Cisco’s website, I went to the product, I went to data centre, I went to servers. And you know, I followed that track. The second meeting, again, why the PDF was in my cache, I could just simply click the URL and when. So, I did the same thing. In the third meeting, you know, my team member asked when it is on cash, why do you always go? I said, you know, that’s where, you know, we we make, you know, that wow, factor is there, right? Oh, we are on website of a company like Cisco. Right. Right. So even, you know, those small things are celebrated. And, in fact, you know, obviously, you know, we still celebrate them now, any any smaller, large vein or alliances or recognition from, you know, the ecosystem outside. So that’s how you know, these small joys are, you know, they look forward to such experiences. All right?
Michael Waitze 23:04
Absolutely. I mean, every little wind is a gigantic movie, and it feels you got to celebrate this stuff. So do you look at the Shopify business model. And I’m not talking about the E commerce thing, but just the platform that they built, and then the way that they’ve given access to that platform for third party software developers to then build on it, but also to give those third party developers just from a business perspective, a way for them to also build very large businesses, right, so that all of the value doesn’t only accrue to Shopify. it accrues to companies like Shogun, this is a year ago now, right? Who raised the money at like a half a billion or $750 million valuation? A lot of it actually invested by Shopify into them? Are you looking at the same type of business model where if you can enable other companies to build onto the app stack, right, that then they can also become big companies in their own right, but let’s still sit on the platform. Does that make sense?
Vikram Gupta and Yatin Kavishwar 23:58
Absolutely, exactly. That is that is exactly the idea and the vision that we have unfortunately, the challenges that because we are working in slightly little more complicated technology from aggregating things point of view. So now we are starting to build those connections in the market where other developers can easily adopt our platform and start building we are starting small, because this is something again that I think this is our next was a geek kick that we are expecting coming in from here, right. So now we have done the building part and we have done the customer connect part. So developer connect part is something that is our next frontier to capture got it. And and exactly we would love to have large companies being built on a cross platform and we just take a backseat and say hey, we have the tools. We have the infra we provide you the customer connect as well. You just build and take the technology today. logical conclusion,
Michael Waitze 25:00
what is the impact of what’s happening at open AI? Like all this stuff, you hear about GPT, the GPT, three anti GPT for the Dali stuff like, what is the impact of all of that research and work that’s getting done? It’s almost like real time research in a way, right? Like, absolutely. And this idea of open sourcing it obviously, it’s kind of amazing. But like, what is the impact of that on you? Do you lean on some of that development as well? Like, what’s the intersection of what you’re building and what they’re building? Is that a fair question? Or
Vikram Gupta and Yatin Kavishwar 25:27
it’s absolutely a fair question. Actually, yesterday, just yesterday, I got my access into chat GPT. And I did a quick demo to my entire team that hey, look, and first question, I typed into it. How can we do AI scale on heterogeneous hardware? That was the first query on Chad GPT, to be honest, oh, wow, there’s been sort of that as well. But having said that, so. So the impact and the intersection that this has is twofold. One, it helps us actually create a lot of tail wind, about adoption of technology in general. From a market point of view, people know that now, we are not talking about a technology which is in the lab, we are talking about something that has that has real impact. And it can actually fundamentally change the way we do things. So it is helping in setting up that flagpole, which everybody can look up to. And it makes our job as a company easier to picture solutions. That is one and second, the research and the activity that is happening in here, we basically the envelope of available technology is expanding as well. Right? So there are GPT, three, now GPT. Four is actually in the works, that is going to be another beast. On the other side, there is stable diffusion, and all sorts of this research is fueling a lot of expansion to all of them are I would say what we call it the genetic, generative AI space where you can generate content, right? So this research and this advancement is also helping in automating several tasks in the interpretation, application of AI where we lie actually. So so how we interpret video, how we interpret images, how we do that for for a given customer is also getting benefited, because a lot of that content can actually help us number one, train better AI, number two, or write better code. So we actually tested GPT for some key concepts that are being used. And it is not, of course, as good a job as a human programmer today, but close or close to probably 80% of that job. And third, very important factor that this intersection is going to help us with is that the new architectures that are being built, right. So this actually validates our platform assumption that there is a lot of shifting ground under the world. And we need to have a platform on top of it, which allows you to leverage the underlying technology in a seamless way. Right. So that actually validates our platforms or OS stance as well, because nobody can actually or nobody should worry about everything that is happening in the world to build a new technology, right? We take care of that for you. You just build on top of the interface that we provide to you.
Michael Waitze 28:12
This gets back to the functional abstraction stuff that we were talking about earlier, right, this idea that if I want to optimise my video AI code, why should I do that myself, Washington, I just talked to them operating system that’s already done the optimization to the machine. But But again, this is not. This is not like a novel concept. We do this everywhere. But what you’re doing is you’re abstracting it out and taking that function and just highly optimising that this should be obvious to people when they look at it, right? Can I ask you this too, though? Are you surprised is like the wrong way to? It’s the wrong word choice here. But like when you started doing this back in 2015, you had this vision of like, where artificial intelligence and its relationship to video and the ability to analyse it could be. But how would you characterise how much the technology has changed? Because I’m not involved in it every day, right? If you know, and I always say like, the guy who designs the windshield wiper on the car must be bummed out, because he’s still like a car engineer. But the windshield wipers haven’t changed much in 150 years kind of thing. But what you’re working on what you guys are working on is changing like moment by moment. And I’m just curious for those of us that are uninitiated how sophisticated and how much different it is today than it was in 2015.
Vikram Gupta and Yatin Kavishwar 29:27
Okay, I think I’ll do a short answer, right. The difference is actually like having exactly on the level of having assembly language or assembly code or Python code. It’s that difference. Actually, yeah. So So what computers Computer Engineering went from probably, I would say late 60s 2010. The journey that was travelled in 50 years, the same journey. AI has travelled in five years. That’s what I want to know. Same level of transition, I think probably more to be honest. Right? So 2015 I I can tell you writing a simple application, the technology, the hardware requirements, the underlying support that was available, right? It was like banging our head against the wall every single second. Now, law, a lot of these challenges have been more structured have been optimised at a great degree and we have changed, or I think one thing that we have done well so far is to add them very quickly as well. Like with newer technologies coming in newer optimizations coming in, we have also stayed ahead of the curve in adopting them before they even get to public in most cases. So yes, it has made our job tougher in one way. It has made it more exciting in another way. But the change is, is drastic, I think being able to write one application was a six month job on a Windows platform in 2015. Today, it is probably less than six days job. So you can also compare it that way. Right? The speed has increased dramatically. Yeah, in all respects,
Michael Waitze 31:05
I think a lot of people use AWS as an example of how a piece of technology is abstracted away, right? And this feels like the same thing in the sense that I remember so you don’t know this. But when I was at Morgan Stanley, I was actually a Unix systems administrator. I went through a course that Yeah, I did. I went through a course at Sun Microsystems in New York and did all that stuff anyway. But building your own server from scratch was something that we did do. We wrote our own scripts, we did the software, we could implement it and propagated across the whole place where we maintain those servers and the desktop machines. And yet one day, AWS just went to send us a check, and you’re already done. And it’s highly optimised, right? But I’m just making the analogy so people can understand because it’s kind of what you’re saying, right? Like, if you want to do video AI, that’s great. build an app for it. But don’t do all this back end optimization, because it’s a waste of your time. And we’ve already done it for you.
Vikram Gupta and Yatin Kavishwar 31:53
Absolutely. Okay. And our experience says the split is anywhere between 10 is to 90 or maybe 20 is to 80 In the best case, right? So 10% is the core effort and 80% is the other engineering effort, right? You it is not your core job at all, no, let somebody else take care of it. And exactly like AWS takes care of managing servers for you. And we will take care of managing the underlying infra for you.
Michael Waitze 32:18
What is the what is the implication or the impact of quantum computing on this space? And how hard is it to port the OS from one hardware architecture or one architecture to another?
Vikram Gupta and Yatin Kavishwar 32:35
Probably an ignorant answer. I don’t fully understand the spread of it. But but from what I understand, quantum computing is great at trying different combinations, which were linearly happening earlier, normal computers, quantum can actually try multiple states at once. That is what quantum computing is all about. So it is great at cracking code and maybe helping in some sort of highly efficient compute. As of now, at least in my belief, the way deep learning or AI software is written, it is a linear network of nodes or a series of nodes that basically process incoming information and give you insight, right, so if we pass in an image, it goes through stages of processing, and tells you there are four people in that image, for example. So in this architecture, mapping, the advantages of quantum computing into it are not that obvious yet yet, not only to me, I don’t think to a community at large, it will need a fundamental rewrite of how this compute is done. And I’m sure people are already working on that, in general, for us, or for this industry to adopt new hardware or new hardware architecture is a big learning curve, always because this fundamentally changes the way of things written in the first place. And then software optimization is done. The second places well, actually, we are proud today to say that we are now portable on Intel, Nvidia, Qualcomm, AMD and a couple other younger acceleration hardware, of course, not quantum computing there yet. So so we have done our bit in making sure that our apps are portable across these five hardware platforms, including kinara. Another very promising hardware platform that we are working closely with as well. But yes, it takes a lot of effort and as an OS. Now it is our job to make it easy for everybody else. When an app is written for, let’s say, windows, right? It doesn’t matter whether it’s going to run on an Intel hardware or AMD hardware, for example, right? Or same happens for an Android app. We are exactly doing the same for video AI, as
Michael Waitze 34:37
well. Got it. Do you want to give any closing comments talking
Vikram Gupta and Yatin Kavishwar 34:40
about one core company core element around in the ecosystem, which is the customer, right? So I think what Wickham mentioned about 8020 or 9010, where 10% is the whole algorithm and you know, the rest of 90% is the engineering around it. And that’s whatever else is doing thing we need to understand what is driving on both sides customer as well as tech customers making significant investment capex investment in cameras in network in buying expensive hardware for inferencing. Right? Obviously, he is not going to just use one application, how do he derives the investment, you know, out of audit ROI, which is he’s trying to what we have seen is from 18. Till now, on an average customer is using at least 10 applications of average loss, on average, on the same infrastructure. Now, if that is the pace at which the adoption of application is increasing, you know, the development of the application also has to match the same pace, right. And that is the exact point where these two requirements and everyone else’s thought process envision meets together, we are trying to enable and accelerate the development of application by creating this ecosystem, and creating a marketplace for the customers to adopt application across domains. And that is where we believe that we will be able to serve each and every customer with whatever applications or field of view or the camera allows you to run, right, not every application will be able to run on every camera field of view. I think this is the interesting equation, which is building up. And we believe that, again, the technologies like 5g and edge AI, these two are going to be very, very important in enhancing and accelerating the adoption of applications by the customers. I think here the behaviour is very important to understand where this whole thing is getting driven from.
Michael Waitze 36:47
Do you want to explain to people why do you want to explain to people why 5g matters? And you said edge AI? Yeah, because there’s a difference here between edge computing, cloud computing, which we should probably explain to people, right, because it matters where you know, particularly on a mobile device, like just how close it is to the stuff that’s actually getting processed, so that the latency between what’s being thought and then what’s getting sent is minimised. Sorry. You want to just explain what that differences, the differences
Vikram Gupta and Yatin Kavishwar 37:09
in the way it is completely architected right? Yeah, there are two ways to bring one is either you bring the complete video onto a data centre and process it and incorrect at the data centre. But that’s too costly. And at times the lag in terms of sending the metadata back and action being taken on that metadata is something which will be very important for few customer take example of traffic,
Michael Waitze 37:34
I was gonna say a car moving at 120 kilometres now I’m moving that
Vikram Gupta and Yatin Kavishwar 37:37
120 kilometres an hour and suppose you know, there was a violation of traffic signal, right? I think the local authorities want you know, the metadata and the action to be generated real at where the action is happening, right, because the, the edge or a traffic signal. So when we talk about edge AI, you know, that will enable faster decision making for businesses. Take example of a foundry, wood chips, or maybe, yeah, not not the chips one, but the real manufacturing, heavy degree. So how Ajay will impact is, let’s say there is a safety compliance issue at a remote factory, and there are connected factories. So it cannot happen that the video travel all the way down to our data centre. And, you know, it will basically delay the process, it may be a safety hazard. So the real time inferencing will allow the person at the action point to take decision immediately. And, you know, this is actually prevalent across a lot of applications of compliance, traffic security, safety. So this is going to impact in a big way. Obviously, one factor, which will also be important to look at is how the cost of hardware inferencing comes down. I think the telecom operators, developers of the hardware, all have to kind of come together and, you know, boost this whole idea of adoption, still, we believe it’s a consumer would take a call, whether to go for a particular architecture, depending upon how much expenditure is getting incurred, and all the capex is incurred in those, this whole lot decision making.
Michael Waitze 39:18
If you’re building your own OS, right, and you’re optimising it for your own hardware, sorry to jump back into this again, but just a little bit more stuff. That’s interesting to me, you’re also going to have to build your own tools for development on that OS, right. So and because it’s a new operating system, the tools are probably also new. In a way it reminds me of what Apple did when they moved from Objective C and Swift. And they were like, This is the new paradigm. It’s not finished yet, per se. So how closely are you working together with your partners in the development, not just of the OS like we need to have this access or that access, putting the tools as well so they can build the proper applications for that platform?
Vikram Gupta and Yatin Kavishwar 39:54
Yeah, so I think I’ll give a very good short answer. We have a tool called Genie. It’s basic CLI not a tool, it’s a tool chain called G, which does the magic behind the scenes for you gives you a boilerplate of a sandbox to play for play around. And you just plug in a few your bits in there and get up and run it lightly. I think more involved answer is that we have taken a conscious choice in not building or not changing the way significantly the things have been done. So for example, if a person is familiar with a certain way of writing, computer vision or video a applications we have used that similar format and programming languages and interfaces, so that it is the transition from how things are done outside of zeros and how are things done within videos. nearly the same. Of course, there is some learning curve, but we have not drastically changed consciously, drastically changed that approach. We have used the existing tools change as much as possible. Of course, we have built our own libraries on top of it, which is hopefully not going to be a steep hill for anybody else to climb.
Michael Waitze 41:08
Okay, boys, I’m gonna let you go. I really appreciate your time today, Vikram Gupta and Yatin Kavishwar who are Co-founders of Awiros. I hope you enjoyed this as much as I did. This was awesome.
Vikram Gupta and Yatin Kavishwar 41:17
This was lovely. Thank you, Michael. Have a great day.
Thank you. Thanks, Michael.