Projects
Things I'm building and have built.
Client platform generating regulated care documents with LLMs. I built the generation pipeline and the reliability layer around it: an LLM-as-judge evaluation harness on a stratified fixed case set, hallucination-rate observability wired to tracing, and resumable eval runs for unattended prompt A/B testing.
An automation layer where coding agents do real delivery work: CI bots for code-quality triage and backlog pickup, guarded so review agents cannot approve their own changes, plus a Slack agent that triages Sentry alerts with cross-repo code and issue search.
A Korean reading app for Japanese speakers: short pieces at your level, explained through Japanese — particles as は/を, Sino-Korean words through their kanji cognates, endings through です・ます. The landing page is live and an LLM pipeline drafts, glosses, level-tags and QA-checks the stories; the app itself is still being built.
This site. A hub for my work, writing, and tools. Next.js App Router, four locales, statically rendered.
Web, mobile, and infrastructure work for teams in Japan and abroad — zero-to-one product builds, Rails-to-Next.js modernization, and AWS infrastructure as code.