Applied LLM engineer — evaluation, retrieval (RAG), and data quality. Building Kyren from Tokyo · ACL 2025 co-author · → USC MSCS '28.
Making "good LLM output" measurable — eval harnesses, production RAG that survives real users, and reward/annotation design built from observed errors.
| Product | What it does | Status | |
|---|---|---|---|
| 🎯 | Rubric | Open-source eval harness for RAG & agent systems. | Building in public |
| 🤝 | KASHITE | Track who borrowed what. | Live |
| 📖 | YOMU | AI document Q&A with citations. | Live |
| ✍️ | Phrasely | Write better English, learn why. | Live |
Python · RAG · LLM evaluation · Next.js · React · TypeScript · Supabase · Claude API · Vercel
- 🇯🇵 → 🇺🇸 Tokyo → Los Angeles for USC MSc Computer Science (Fall 2026).
- 🎯 Building Rubric — an open-source evaluation harness for RAG & agent systems.
- 📄 Co-authored at ACL 2025 (Enhancing AMR Parsing with GRPO) — data & evaluation.
- 🤿 PADI Master Scuba Diver & Rescue Diver · 60 logged dives.
kyren.app · masatonaut.dev · LinkedIn
Strip away the unnecessary. Ship what matters. 簡素 (Kanso).