Ashmit Bhatnagar: From UT Austin to AI

Picture a college kid in Austin, Texas. On his resume, in the same twelve months: live communication software for a rocket, software for a mining vehicle that digs tunnels, and then a job at an AI company valued in the billions.

That's not a resume, that's four people's resumes stapled together.

It's one person. Ashmit Bhatnagar. And the question I kept circling is... how does somebody go from undergrad electrical engineering to the actual frontier of AI? Because the answer isn't one lucky break.

Okay, but before we build a legend here — how much of this is actually public? I've seen people inflate a LinkedIn.

Fair, and I want to be upfront. What we have is his own professional record and his GitHub. There's no big profile piece, no interviews. So this is a story told from the paper trail, and I'll stay on it.

Good. Where does the trail start?

University of Texas at Austin, Bachelor of Science in Electrical and Computer Engineering. And almost immediately, January twenty twenty-three, he joins Texas Guadaloop.

Guadaloop as in... hyperloop? The vacuum tube thing?

The student hyperloop team. He's an electronics engineer on vehicle systems for a year and four months. So as a UT undergrad, he's wiring up the systems of a pod meant to move at serious speed.

Meanwhile my college extracurricular was intramural dodgeball.

And then he does the thing that I think is actually the hinge of this whole story. January twenty twenty-four, he becomes a teaching assistant in the Cockrell School of Engineering. And he doesn't do it for a semester. He does it for a year and seven months.

Teaching assistant. That's the hinge? That's the least glamorous line on the page.

Look at what he taught. Algorithms. Software Design Two. Honors Software Design One. Honors Embedded Systems. That's the hardest core of the degree, and he's teaching all four.

Huh. Okay, so you're saying he spent nineteen months explaining algorithms to other people.

Which is the fastest way to actually own a subject. You can fake understanding in an exam. You cannot fake it in front of thirty students who are allowed to ask why.

That's a genuinely useful thing. Teach the thing you want to be good at.

Then twenty twenty-five goes vertical. January: he's a software engineer at the Texas Rocket Engineering Lab, working on rocket live communication. Telemetry. The stuff that tells you what the vehicle is doing while it's doing it.

And that's a semester? Four months?

Four months. And in March, while that's still running, he's a Founder in Residence at Afore Capital — a pre-seed venture firm in San Francisco. His project: multimodal AI evals.

Hold on. He's a VC-backed founder-in-residence while he's still a student TA?

Yes. Afore runs a founders-in-residence program, invest-in-you-on-day-zero. Their whole posture is build now, raise later. He spent three months there on AI evaluation.

Evals being... how you measure whether a model is actually any good?

Exactly that. And remember that word, because it comes back.

Oh, you're setting something up.

September twenty twenty-five, he's a software engineering intern at The Boring Company in Austin. Mining vehicle software. The tunnel-boring machines.

From rockets to literally underground. That is a weird pivot.

And simultaneously a research assistant at UT in the GIS lab — geographic information systems. Mapping, spatial data. Which, if you're writing software for a machine chewing through the earth, is not a coincidence.

Okay, he's stacking jobs that secretly rhyme.

That's the pattern. Hardware you can touch. Systems that have to report where they are and what they're doing. And then — June twenty twenty-six — San Francisco. Member of Technical Staff at Mercor. Evals infrastructure.

There's your word.

There's the payoff. The thing he built as a founder-in-residence is now the thing he does at a company that TechCrunch reported raised a hundred million dollars at a two-billion-dollar valuation in February twenty twenty-five — founded by three twenty-one-year-old Thiel Fellows out of their dorm rooms.

So the student engineer ends up at the company built by students.

Mercor's pitch is organizing human intelligence to power the AI economy — RLHF data, agent training, for the top labs. Which means Ashmit's job is building the measuring stick for AI. And his own stated interests? Mechanistic interpretability, inference deployment, physical AI.

Interpretability being the "why did the model say that" problem.

The hardest open question in the field. And he got there not by picking AI at eighteen and tunneling straight toward it — but by doing hyperloops and rockets and dirt-moving machines, and teaching the fundamentals until they were bone-deep.

So the lesson is don't specialize?

The lesson is specialize in the layer underneath. He never chased a job title. He chased systems that have to work in the real world, where the feedback is brutal and immediate.

And the TA thing. I keep coming back to that.

Because it's the free move. Nobody can stop you from teaching what you want to learn. Everything after that was just picking harder machines.