About

I’m a computer vision engineer in roughly the way anyone is whatever their company needed most. It’s what Athlitix requires, I’m good at it, and I enjoy it. The less tidy version is that I’m curious about almost everything and vision is where the curiosity currently happens to be pointed.

The list, if you want it: maths and physics, LLMs and agent frameworks, ordinary software engineering, DevOps and automation, networks, computer architecture, embedded systems, microcontrollers, IoT.

None of that was a plan. I picked things up because they looked interesting — in a course, in a project at two in the morning, at a meetup I went to for the free pizza — and then every single one of them turned out to matter when I started building a company. Not one has gone unused. I’m aware of how unlikely that sounds.

I think it’s because a specialism is the top layer. Underneath it, building the thing and then getting it into someone’s hands demands the whole stack: the maths, the machine it runs on, the network it crosses, the pipeline that ships it. You can be a specialist and not know any of that. You just can’t be a specialist on your own.

What I actually enjoy is when two fields turn out to be the same field. The control systems I learned for motors in undergrad are, structurally, reinforcement learning. The signals-and-systems and DSP from an electronics degree keep resurfacing in vision tasks, usually in ways the textbooks did not have in mind. Most of my better ideas are something borrowed from a subject that wasn’t supposed to apply.

The distinction I’d actually argue about is a project versus a product — that one got long enough to be its own piece.

Athlitix

CTO and co-founder. We’re trying to make performance analysis something anyone can do from a phone — no wearables, no markers, no instrumented facility. RinkUp, our first product, does it for figure skating, which is a deeply unfriendly place to point a camera. It’s on the App Store; swimming is next.

It’s built for athletes as much as coaches, and a surprising amount of it is plumbing rather than models. Coaches record constantly and drown in footage, so the video lives in the cloud instead of on a phone that’s out of space. Sharing each clip with each athlete by hand eats an evening a week, so a coach tags a video and the right person just has it. None of that is clever. All of it is the difference between something people admire and something people use.

In 2025 it won the $75,000 grand prize and the audience choice award in the digital technology track of the NYU Entrepreneurs Challenge, after the J-Term Startup Sprint.

None of which happens without Sachin Dasari, who co-founded Athlitix with me and runs it as CEO. He studied finance and entrepreneurship at NYU Stern and played D3 volleyball there, which matters more than it sounds like it should: he has been the athlete on the other side of the camera, and a good half of what we build is obvious to him and wasn’t to me. I do the engineering. He is the reason there is still a company for me to do it in.

Before this

An MS in Artificial Intelligence at Northeastern. Before that about two years at TransUnion, which started as an internship and turned into the job. Before that a B.Tech in electronics and communication engineering at VIT Chennai — the degree that keeps quietly paying off every time a vision problem turns out to be a signals problem.

At the end of it I co-wrote a chapter on machine learning in genomics for Data Science for Genomics (Elsevier, 2023), on identifying and modelling anticancer peptides. It’s a long way from ice rinks and I still think about it.

Somewhere in there: internships in embedded systems, in Linux and DevOps, and as a software engineer at a startup. A stint doing marketing and graphic design, which I don’t regret and which explains why I have opinions about typography.

How I work

I learn by getting my hands dirty. Reading about a thing gives me the shape of it; building a bad version gives me the rest.

I cannot hold syntax in my head and I gave up being embarrassed about that some years ago. I know what I want and where it lives. The docs can keep the details.

And I have never once had enough CPU or GPU. I’ve had considerably more and considerably less over the years, and I can report that the feeling is identical. I assume it would survive a cluster.

Elsewhere

GitHub · LinkedIn · Google Scholar · Medium

Email is hi@girishkumaradari.com.