From identity and payment products to AI products.
01KFTC
One palm, enrolled once at a bank, clears the ID check at 14 airports. I led the go-to-market and the rollout with 12 banks: 500K users in the first year, with zero security incidents.
A biometric payment system that runs without a card network, shipped in one month with zero false charges.
02Recent
Projects from 2025 and 2026, built at the University of Washington, at PKNIC, in hackathons and on my own.
I'm creating a knowledge graph to notice when the rules your agents run on have changed.
With the Office of Research Information Services, I led product and UX research for an AI-assisted redesign of SAGE, the university's research budgeting system. The AI copies and pre-fills budget data, a grant manager confirms each value against its source, and submission and sign-off stay with a person. The redesign is projected to cut grant manager workload by 25 to 33%.
A live LLM app for finding laptop-friendly cafes around Seattle, with a published evaluation.
An internal content tool that writes posts in English and Korean and scores competing LLMs on quality, safety and token cost before any draft is used. In the pilot it cut the work from 30 minutes to 5.
I gave an AI agent write access to my notes, and code checks every write. Handwritten pages, screenshots and saved articles land in one archive with study notes and a quiz: 107 captures verified, none failed.
I won the Data Track, at the Beginner level, in the Humane Intelligence Bias Bounty Challenge. I labeled 55 voice clips twice and compared my labels with an emotion AI model's. Our 74.6% match turned out to be shared bias, not accuracy.
A multimodal agent that suggests outfits from photos of your closet, your mood and the weather. Submitted to the Amazon Nova hackathon.
An interactive story that follows an AI agent through nine moments of its day, and the security check each moment needs.
A Chrome extension that turns a YouTube talk into work insights and the English used to express them. It runs on the model you bring: Gemini, Ollama, OpenAI or Anthropic.
As project lead, I ran the research with a blind student who works through calculus by screen reader. Together we built a recall tool: a keyboard shortcut speaks back any earlier term, so he keeps his place. He rated it 9 to 9.5 out of 10 against a human scribe.
I scored Clearview AI, Face++, AnyVision and Trueface on 28 governance metrics, from disclosure to redress. On human oversight, only Clearview AI described a person reviewing its matches.
I built a desk dock that senses when a phone is set down on it and starts a visible focus timer. A light sensor with a calibrated threshold ignores flicker.
If your team is building products where AI acts in critical roles on a person's behalf, I would like to hear from you.