The Iferia Techcast
00:00:02 · Ezekiel Iferia, Krishna Kumar Sharma
[00:00:02] S0: Welcome to the Iferia Techcast, where we bring you all things, science, engineering, technology, and innovation. I'm your host, Ezekiel Iferia. And today, we are joined by Krishna Kumar Sharma, the former head of AI engineering at Amazon and a seasoned tech leader with over twelve years of experience building large scale AI systems. He recently left the corporate world to found Omukai, a sovereign AI startup building multimodal interfaces to control robot drones and machine swarms through natural language. He's also the author of the upcoming book, Agentic Coding: The Boring Way. So, Krishna, I am really honored to welcome you to the show, sir. Thank you very much for joining us here today. [00:01:01] S1: Thank you so much, Ezekiel, for the invitation and then a kind intro. [00:01:08] S0: Alright. It's it's an honor to welcome you. So let's get started, Krishna. You you were the head of AI engineering at Amazon, you know, one of the most powerful tech companies in the world. So you you left to start Omukai, a sovereign AI startup. So what does sovereign AI mean in plain English? You know? And what made you walk away from a top tier corporate role to be something of your own?
00:01:36 · Krishna Kumar Sharma
[00:01:36] S1: Yeah. The... So first thing about Sovereign AI. Sovereign AI basically means that you want an AI system which is not controlled by any other country. Basically, something which you can deploy yourself. So I'm based out of Germany, so I'm making sovereign European. And in this context, it means that the tech is built in Europe and primarily for Europe, and then we cannot have any sanctions or controls from either US or China or something like that. So example, like US cannot tomorrow go and say, Hey, stop delivering your services to Germany. But because we're German based, US or China cannot control what we are doing. This is where the concept of sovereignty comes in, that you are able to build for the country or the region of your choice, and then without any control. In terms of why I decided to leave my job, it is... How do I put it? Amazon is a great company, and I learned so much over there. In the last five years, I've worked across robotics. I've worked across government clouds, a lot of interesting domains. But if you see in the last two to three years since OpenAI and Anthropic came in, Amazon is, I would say, in terms of the AI race, not doing that great. I think Amazon is like a prime investor in both Anthropic and OpenAI, so they actually have stakes in the AI. But in the engineering, like as someone with a lot of history in engineering, I realized that I would like to solve more complex problems as my career is progressing, and I was not getting the opportunity to solve those particular problems in Amazon. And the way I decided to do that was that actually I left Amazon and I joined another corporate company in Germany, and I tried to balance my startup with this company in parallel. But after a few months, I realized that, oh, this cannot work. Mean, if I have to do a startup, I have to go all in, and trying to balance a job with a startup is not possible. So I left this corporate company altogether, and I jumped into my startup 100%. And the same system cannot work in Amazon because of the different regulations Amazon have, the contractual, etcetera, which did not permitted me to, you know, take a big step.
00:04:09 · Ezekiel Iferia, Krishna Kumar Sharma
[00:04:09] S0: Alright. Thank you very much for sharing that with us. So, Krishna, you know, I'd like you to, you know, give us an overview of what you are building at Omukai. You know, said a a word or two about it in the introduction, but I'd like you to maybe just walk us through the the the, you know, the the the the tech you're you're building, tech you're working with on a day to day basis at, you know, Omukai. [00:04:39] S1: Yeah. So in simple words, maybe I ask you a question before I explain this. Right? Have you ever... What... When you think about, like, you know, controlling a robot or a drone or any machine, do you think it is easy, difficult, you require training? How do you see this process as you and person who has never worked with any such robot or a drone, and now tomorrow, I just give you one robot? And then how do you think will be your interaction or how you will manage it? [00:05:11] S0: Oh, you know, if I'm someone who has never worked on, you know, similar projects, you know, operating a robot or a drone, think it's gonna be a hard time for me to, you know, get it right. [00:05:24] S1: Yes. Exactly right. But imagine you can just... But imagine if you had another human being you have never met before, would have been it easier for you to communicate and tell the human what do you want to do or get accomplished? [00:05:38] S0: Absolutely. Absolutely. I mean, it would be much more easier, you know, communicating to a human, you know, what I want the person to do than to, you know, do that to a robot. [00:05:50] S1: Yes. Now imagine
00:05:50 · Krishna Kumar Sharma, Ezekiel Iferia
[00:05:50] S1: Yes. Now imagine [00:05:52] S0: add, you know, prior knowledge about a robot. [00:05:55] S1: Yeah. Exactly. And now imagine, you can just speak to a robot like you speak to a human, and you just give a natural language command that, hey. Go to that particular spot behind you and bring me that glass of water. Imagine if you just give a simple command and the robot is able to understand you and do the action. Now this is exactly what we want to do. We want to create a world where you use our tech, and you can speak to machines like you speak to another human. This is the very simple way of what we're trying to do. And the way we're trying to do is that, instead of just speaking to one robot, now think like you have five robots doing different tasks, and you have five drones doing maybe shooting a show or doing security, etcetera. And from a single interface, single screen, you can command and control all of them, just selecting them and then giving a command in your voice. This is the future which we are building.
00:07:00 · Ezekiel Iferia
[00:07:00] S0: Wow. That that... That's quite fascinating. That's interesting. So, Krishna, you know, you are building a multimodal AI interface that lets people control robot drones and machine swarms, you know, using natural language, you know, like communicating to a human. So in... Rather, how how close are we, you know, to being able to just talk to a robot and, you know, have it do what we see? I mean, what's what's like the timeline into the future we are looking at for that thing to be like, you know, the main the mainstream technology and, you know, anyone simply can just get a robot for the first time and give it instruction and it actually delivers?
00:07:45 · Krishna Kumar Sharma, Ezekiel Iferia
[00:07:45] S1: I would say like we are not very far away maybe. So basic task, can do it today already. And maybe a bit more complex tasks, it will require one to two years more. So the tech which we have built, we are already able to command robots through very simple tasks like going across a parameter, trying to scan the area, trying to send the signals, etc. Very basic things are already working out. And now if you want to do a bit more complex task, example, you have a robot and you want this robot to act like your personal butler, and then this robot should be able to do multiple tasks, right? Like, I don't know, fold your clothes or bring you a glass of water, all these different kinds of things easily. Currently, like, you have to teach robots one individual task at a time, and then giving all the task... Different task at once is still a bit difficult. But the way we are progressing fast, I think in one to two years, it will be possible that you have a robot which understands hundreds of different tasks already and can... And is able to perform all of them as per your wishes. [00:09:07] S0: Interesting. So, you know, we do see videos, you know, of companies coming with humanoid robots to do, you know, a demo, and we see them give the robot instructions and they start maybe showing their dance moves, start doing back flips, and all of that. But can you give us a specific example of maybe a military or industrial use case where this would actually be a real game changer?