About

Last updated: 2026.07.14

I'm 19, from South Korea, heading into the Army soon. I'm interested in
startups — and in eventually founding a research-based one.

Background

KSA. At the Korea Science Academy I got deep into BCI and Brain-Inspired AI —what pulled me in wasn't just the technology but the questions underneath it: the self, cognition, why we do what we do. I spent two years as a student researcher in a lab at KAIST (Prof. Sang Wan Lee), working with fMRI data and machine learning, and led an R&E team. I've since moved on from that field, but it's where I learned how to actually do research.

KAIST. Studied here after KSA; currently on leave. For a while I chased quant — did some alpha research on the WorldQuant platform — because I thought building capital was the prerequisite for building anything big. But the more I looked at how the money actually got made, the less it moved me: it's either arbitraging inefficiencies or winning a zero-sum game against the other side of the trade. Neither felt like creating anything. I wanted to play a plus-sum game — to earn by putting something new into the world — so I walked away.

Altos Ventures. Altos is a US venture firm managing more than 10 billion dollars, known for early bets like Roblox. I joined as a Hacker-in-Residence in May 2026, scouted by one of the GPs to build out his AI workflow — turning the way he worked into tooling.

EO House. A month at a founders' hacker house in San Francisco, June 2026 —met a lot of people, ran a lot of coffee chats, tested a lot of half-formed ideas.

What's next

I'd like to build something of my own, on a research foundation. Right now that
means a research proposal I'm developing, ValueRank — a rethink of how ranking systems assign value to information.

Today's feeds rank by engagement, a proxy that drifts away from whether
information is actually good for you, and they infer what you want from your
behavior rather than letting you say it. ValueRank inverts this: the user
declares a goal, and items get ranked by the change they're predicted to cause in the user's own state along that declared direction — making your stated goal the unit of value, not a filter bolted on top of engagement. Because value is a projection onto that goal, the change a system induces in you becomes auditable rather than assumed away. First testbed is language learning; the open questions are how far it generalizes beyond it, and whether the machinery beats simply prompting an LLM with your goal.

Entering military service on 2026.07.27; back in about 18 months.