In The Arena by TechArena - Data Insights at OCP: DataraAI on Edge Robotics

Episode Date: January 16, 2026

DataraAI CTO and co-founder Durgesh Srivastava unpacks a data-loop approach that powers reliable edge inference, captures anomalies, and encodes technician know-how so robots weld, inspect, and recover like seasoned operators.

Welcome to Tech Arena, featuring authentic discussions between tech's leading innovators and our host, Alison Klein. Now, let's step into the arena. Welcome in the arena. My name is Allison Klein. We are coming to you from San Jose, California at the OCP Summit. And it is a Data Insights episode, which means I'm with Janice Haraspi. Welcome to me. Oh, thank you, Allison. It's great to be here. We're wrapping up day two of recording at OCP Summit, and it's been a wild ride. Janice, tell me who we've got with us right now and why this is such a unique episode. This is a very unique episode, and I'm very excited about it because we don't get to talk about this much, but we're going to talk about robotics.

Thank you. Yeah, I'm excited. Thank you. Yeah. So today we have to Tara AI and with us, Rakesh Zervasta. Yeah. And you are currently the CTO. Yeah, and founder.

Yeah. And chief, the whole bit. Yeah, exactly. So let's just dive right in. We love to learn your perspective on what you're doing with Deteri AI. And tell us about robotics. Yes. So what we're doing is we want to make sure the industrial robot,

especially for manufacturing, like cars or heavy machinery, they are accurate. So that's our problem statement. And what we're trying to do, this is a big market, $30 billion in the next three years. What we're trying to do is provide the right data so that the models are trained properly. they can differentiate in welding versus other activities. That's what we are focused on. We want to change the industry 4.0 and help USA accelerate industry 4.0.

Degas, you've been in the industry for a really long time, and I know that you understand the compute landscape incredibly well. When you look at the industrial robotics space, where do you think of it in terms of its maturity as compared to some of the other areas of compute? That's a very good question. So if I look at the data center, if I look at automotive ADAS, or the manufacturing, it is a lot more mature.

Robotics is still growing. And thanks to Nvidia, there are a lot of hardware and software solutions. But it still, it is in fancy in the earlier stage. And that's where it is important that we work on it and make sure we can build the dark factories or industry Ford Auto. So it's a good question. It's very early. Yeah.

So with it being very early, right, because robotics, general is very early, I think at large. We still have a long ways to go. What are some of the challenges that you're seeing in terms of building up an infrastructure within robotics? Yeah. Another very good question. The main challenge right now is the data. So when I start looking at it, talk to various mec's and friends, the problem is the data is not accurate. And if data is not accurate, the models won't behave accurately as well. The first problem given by one of the top automotive manufacturer was that can you take our data, which is a static data, and convert it into a robot IVEU, what robots are for welding or windshield or

bumper. So that's one of the big challenges right now. And once we solve that, then the robot can do multiple tasks, not focus on just one task. How do you see the AI challenge in this space across traditional ML applications and some of the more advanced models? Yep. Thank you, Jason. So So before the large language models, of course, the computer vision, robots have these cameras, sensor, just like car, and they were able to detect. But with large language models, what we have done is we got a lot of static images for the manufacturing, and we can use a large language model to construct a 3D from a 2D images. So whatever is the missing portion, we can generate it.

Oh, that's very cool. And very accurately. So now you can imagine, I got a 2D image for a car, and I can generate and put the camera anywhere. So if one car manufacturer wants camera at certain level and other ones a different level, we can place it because now we have 3D. So that's what we are building. And because of AI and large language models, that has become very effective, very accurate. And we have a lot of things like something called as part packer.

So when you're manufacturing car, when you put the door on it, the shape change. Because if door is open, the shape changes. So part packer, it is taking various parts, components and packing. it together. So then even if shape changes, it can figure out the 3D. So there are a lot of interesting things because of the way I. So there's a lot of buzz out there. Oh, AI, you know, robotics to take away my job. Are you seeing less folks on the assembly line and more of robotics just taking over? Not right now. That's the go. It will happen. The reason is again, we have the labor shortage, of course. And then to be competitive in the international market,

that we have to have these dark factories, automated factories. So right now it's not, but that's the trend where right now a robot will do one job like nutball, rather welding. So you shift the car from one station to other, and that creates the inefficiency. But the goal is you have a robotist can do multiple tasks in making more efficient.

And yes, at some point, yes, the human in the loop will be less and less. Are you able to manufacture a vehicle more quickly as a result of implementing more robotics? That's exactly right. That's the goal. So if you can even shave off like 15 seconds, that results in millions of dollars of additional revenue for car manufacturers. That's amazing.

Now, you've talked about manufacturing of automobiles, but that's not the only application here. Can you talk about some of the other targets within the manufacturing arena where robotics is getting advanced? Yeah, good point. So other thing which is very dear to my heart is the silicon area. I have worked at Intel, Nvidia, and MIPS, and now my own company. So in the silicon manufacturing, there is a lot of automation which is required in the manufacturing process.

It's starting from the way for sort and going all the way to the packaging. So I do see the same technology or platform we are building. The next one we will target is the silicon manufacturing. Right now, we just want to keep this very focused, which is the heavy machinery and car. but the same platform building it a way which can be applied to other places. That's amazing. Yeah, when we think about how all this comes together, whether it's hardware or software, what are some of the software tools that you're really relying on today in the early stages to build this up?

Great point. So what I've learned over life is that you can build best hardware, but unless there's a good software, it fails or it doesn't work. With my Nvidia background, I'm using a lot of Nvidia software. So what Nvidia does is not just the hardware, but Nvidia has a simulation platform. And you hear digital twin from Nvidia. So these platforms and these softwares are providing me

ability to simulate a industrial environment or simulate, like Part Packer, which I said is based on those tools. All this is based on very good software, which is either simulation, model generation, or the action for the planning stage. So it's all beautiful software, which we are putting it together to build this data. ecosystem.

Yeah, this is a huge, in your company is named Atara for it, this is a huge data problem. Yeah. And you're doing a lot of applications at edge. If we go back into the infrastructure, how do you tackle delivering these types of very data-rich applications? You know, are you training at the edge? Are you training centrally and then moving to inference at the edge? How do you approach that?

Very good point. So for the training, we are still dependent on the back. bone of the data center because we are not real-time training. You're collecting the real-time data. So the edge is really the influence, which is looking at the object, recognizing, and if robot is welding and by chance there is a reflection or bright light, it can still do it. So those are the inference, which is at the edge. And data we keep in the back end and daily, we update the data. Very nice. What are you kind of most excited about within the next two years? What do you think is really going to progress next for robotics?

It's the full automation. The robot can do multiple tasks as good as human. So another thing which I observed were talking to the industry people is that the technicians are doing the job. They have 15, 20 years of knowledge, tribal knowledge in their brain. So if there is anything which happens, which is not according to the plan, they know what action to take. So those kind of things, just bringing that knowledge into the robot I'm very excited about. Now, Degesh, you know, I've got one more question for you. Obviously, Detourai, it's a new company, startup.

What has the market response been thus far? And what are you looking forward to as you head into 2026? Excellent point. Just like any startup and being realized, we talk to multiple industrial and robotic experts. So everyone agreed the data is a problem. Some of them brought up some of the points that how do you use? make the data so that it's applicable to various companies and various sensors. So we have addressed that problem.

But overall, as we go, in general, the industry response, the investor response has been very positive, that this is the fundamental problem. A lot of people are working on algorithms and robotics and humanoids, but there's less of the people providing the data. So response have been phenomenal and we continue to work towards it as we go into 2026. our plan is to deploy at least to products at two companies, and then we go from there. Nice. I think a lot of folks are going to be very interested in your organization.

Thank you. Where can folks go to learn more and get in contact with you? Yeah. So right now, our website, of course, is the first place, data.a.a.i. And of course, LinkedIn, I'm very active on LinkedIn, which is Durgeshiwastwa. You can find me on the LinkedIn. and email is Durgish at dataraa.ai.ai. So all the means, I welcome all the questions.

It does not have to be positive. I'm very often. All the negatives also, you learn from me. Yeah. That's fantastic. Well, Durga, based on your track record, I have a feeling there's going to be a lot of thoughts of it. Thank you, Eddie.

So let's have you back as soon as you've got some word to share. Yes, and demo or something. Yeah, definitely. And we thought we were rocking another episode of Data Insights. Janice, thank you so much for the partnership. It's been great. And Gash, thanks for being on the program. Thank you. Thank you.

Thank you. Thank you. Thank you, Jenny. Thanks for joining Tech Arena. Subscribe and engage at our website, Techorina. All content is copyright by TechRena.