Seattle ↔ New Delhi
I find what hurts, then build at it.
That’s taken me to oceans, brains, street crime, protein structure, and AI agents — so far. The fields don’t repeat. The reason does.
Right now: CS + Data Science at UW (’27), cryo-EM protein structure in the DAIS lab, and voice + email agents with guardrails even I can’t talk my way past.
what I’ve built at
- Taught 350+ kids computer science and AI — SkillTern, co-founder and CTO.
- K-means over 10K+ LAPD crime points, SOS in Swift, into Microsoft for Startups — then I shut it down on purpose.
- cryo-EM density maps → 3D protein structure, in UW’s DAIS lab. Right now.
- 20+ classes of bug caught at 89% on the repos I tested — shipped as a VS Code extension.
the filter
I don’t take small problems. Oceans, brains, street crime, access, agents — every one of them was bigger than me when I started. That’s the filter, not an accident.
And I don’t walk in without an edge. If I don’t have one, I go build the thing that gives me one: the eval battery that attacks my own work, the model panel that judges my own words, the tool that reads the code before I do.
the move
- 01 problem Something that actually hurts someone — a traveler getting robbed, a kid locked out of CS, a paralysed brain, an acidifying ocean.
- 02 prototype The most ambitious thing I can build at it with what I have. Ships rough before it ships polished.
- 03 teach / give away The point was never to own it. Teach it, open it, or let it die honestly when it should.
exhibits — each one the same move, different medium
-
The clearest version of the whole pattern: find the people a field forgot, then hand them the field. Curriculum, website, certification system — all of it built so the thing could outlive me, and then I stepped off. That was always the point, not the exit.
-
A tourist-safety app I co-founded: flag the blocks where people actually get hurt, then route you by what you came for — sightseeing or speed. The ending is the part I keep: killing it cleanly was harder than launching it, and more useful.
-
The current chapter. A voice-and-email agent handling first-contact recruiting end-to-end, and the guardrails around every action it can take. It runs on a synthetic world, so the stamp reads Demo, not Live. The agents are the medium; how and why I run them is the point.
-
The craft under the agent work — a.k.a. “manipulating the AI.” Getting reliable, exact, repeatable output from a people-pleasing text predictor: personas, hard rules, output contracts, and evals built to attack rather than flatter. Even the headline on this page was steered, not typed.
-
A cabin-design platform where the generator is the easy half. Around it sits a Seattle building-code rule engine, multi-objective optimisation that keeps the trade-off instead of burying it in a weighting, full plumbing and electrical routing, and export to the formats a plan reviewer actually opens. Generative design demos beautifully; this one has to survive the rules that decide whether a building can exist.
-
Optogenetics, brain-computer interfaces, connectomics, and building a CNN from scratch — written up as I learned them. Not lab credentials; the autodidact’s habit of explaining a hard thing to make sure I actually understand it.
-
Real research plumbing in Dong Si’s lab: the unglamorous passes that feed the pipeline, and renders a structural biologist can actually argue with. Teaching a machine to read structure out of a blur is the same move as everything else on this list, pointed down a microscope.
-
Before any of it was a career: H2OAquatics (an AUV against ocean acidification), and civic platforms aggregating NGO campaigns. The instinct to build at a problem started young, and it started civic.
The door
That’s the short version. The long one — what I’m actually chewing on right now — is the part of this site I won’t fake, and it’s still being written.
Send the cold email → — about the robot, the clean kill, the kids, or the music I haven’t released. The strange ones get answered first; worst case, a fast, honest no.