Projects

PROVEN — Co-founder and Operator. AI personalization that learns what works for each person and improves with every interaction—not a static quiz. Started with skincare, expanded to other categories, scaled to $150M+. The same approach applies anywhere there's high variance in preference and high cost of trial-and-error. YC W18, 2 patents, MIT AI Idol.

Physics research — World's largest peta-flop simulation on 163,840 cores. 10 papers, 2 PRL covers. PhD at USC, postdoc at Stanford.

Earlier — Camera360 (top-10 Android, acquired), Divine Might (top-10 MMORPG globally), Cubicman (top-10 iOS, acquired).

Side projects — Durin (market intelligence infra), Pawgress (family AI).

Durin

Market intelligence infra. Tinkering with semantic grounding for preference data.

Pawgress

Family AI that knows when to step back. Exploring what happens when you don't optimize for engagement.

PROVEN

Co-founded with Ming. AI personalization that learns what works for each person and improves with every interaction—not a static quiz. Started with skincare, expanded to other categories, scaled to $150M+. The same approach applies anywhere there's high variance in preference and high cost of trial-and-error. YC W18, 2 patents, MIT AI Idol.

Noteworthy — AI fragrance personalization, same thesis as PROVEN.

Lyra Health

Employee #28. Clinical ML for therapist matching. Joined at 18 people, left at 200+.

McKesson

Drug pricing and supply chain ML. The data here is what got me into healthcare—#2 and #3 most distributed generics were antidepressants and antipsychotics.

Early Apps & Games

Built during grad school, while doing the physics PhD.

  • Camera360 — top-10 Android app, acquired
  • Divine Might — top-10 MMORPG globally
  • Cubicman — top-10 iOS game, acquired

Research

PhD at USC, postdoc at Stanford. World's largest peta-flop simulation on 163,840 cores. 10 papers, 2 PRL covers.

Stronger and smarter materials

Sulfur-induced embrittlement in metals, Ni-Cu-P/CNT composite coatings. Explains how tiny impurities lead to metal failures; helps design safer alloys.

Nanoscale structures for electronics

Stacking faults in GaAs nanowires, twin superlattice formation, exciton transport. Practical guidelines for defect-free nanowire fabrication.

Clean energy with AI

ML-driven perovskite solar cell optimization (reached 18.5% efficiency in <100 trials), DeepSolar mapping (all U.S. rooftop solar panels using satellite imagery).