A line through everything I've worked on — and, where I can name it, why. It starts with a dead language and ends, for now, at OpenAI.
I started learning Latin in middle school and never really stopped. Around the same time, NCSU and UNC let me sidestep the usual rules to take math classes well above my grade — first in elementary and middle school, then through high school.
I'm genuinely grateful to both for bending those rules. Being handed real mathematics early, before I knew what was supposed to be hard, shaped everything that came after.
I did my first real math research in high school at UNC with Jeremy Marzuola, working in and around the geometry of manifolds through the Manifolds RTG.
It was my first look at how mathematicians actually think — slow, careful, and obsessed with the right definition. I was hooked.
Freshman year, I realized I might be taking my last Latin class ever — and I couldn't accept that. So, naively (I didn't yet know what a loss function was), I set out to build a Latin translator.
The translator itself is nothing special, and honestly I should do a far better job maintaining it. But the point was never novelty. It was friction: making it easier for historians and Latinists to do their work.
That turned into a forthcoming piece with John Martin arguing something bigger — that there's a vast 'continent' of European knowledge (as James Hankins put it) locked away in Latin, a language fewer and fewer people can read well. AI could change that, and in doing so, revolutionize how we do history.
The same year, I joined Duke's Math+ summer program, working with Ilyas Khan and Alex Pine (and a few others from the program).
I also did mathematical-biology work with Jacob Martin around then. Different domain, same pull: using mathematics to make sense of something messy and alive.
At Duke I got into machine learning research proper, working with Dhingra, Larry Carin, and Vahid Tarokh. One was a group project we pulled together in a couple of weeks — Tarokh liked where it went, but none of us had the time to carry it past the class.
Later I worked more closely with Xiang and Carin. This is where the AI thread of my story really took hold.
Professionally, I joined System2 in New York, working on system-2 thinking alongside an excellent team of data scientists.
I spent some time on the side reading and writing data at Outlier (a Scale AI initiative), to get a better feel for what kind of data people were really buying and selling.
I also tossed a few math problems into HLE — ones I found genuinely tricky at the time, though they're probably saturated by now.
Today I'm a Member of Technical Staff at OpenAI, still chasing the same question that started with a Latin dictionary: how do we get machines to genuinely reason — and how do we point that at the knowledge that matters most?