01 / About
A bit about me
I design and ship AI-native products end to end. My workflow is spec-driven — I architect systems in plan mode, implement with Claude Code multi-agent teams, and validate with automated verification loops before anything ships.
Right now I'm building CaseFill, a production AI platform for immigration attorneys powered by Claude. Before that I built self-improving trading agents, computer vision apps, and NLP annotation tools. Every project starts with the same question: what can AI make 10x better?
At a glance
02 / Projects
What I've built
AI-native products — from agentic LLM systems to self-improving trading algorithms to on-device computer vision.
CaseFill
AI case management for immigration attorneys. Agentic research with Claude tool use, document OCR via GPT-4o Vision, form autofill with structured output, and streaming AI chat — all with circuit breakers and PII filtering in production.
Stock Agents
AI copy-trading platform with 15 self-improving algorithmic agents. Thompson Sampling selects the best performer, agents evaluate their own signals and adapt parameters autonomously via EWMA feedback loops.
VisionForm
Real-time workout form analyzer with on-device AI. Uses Apple Vision Framework for 3D pose detection on iOS 17+, scoring exercise technique frame-by-frame with zero cloud latency.
LinguaLabel
AI annotation platform for low-resource languages. Accelerates NLP dataset creation where training data is scarce, helping researchers build models for underserved languages.
03 / Readings
Books & essays I recommend
Pieces that have shaped how I think about engineering, technology, and building things.
- The Idea Factory: Bell Labs
- Conquering the Electron
- iWoz
- Modern Software Engineering
- The Clean Coder
- Moore's Law for Everything
- Ten Lessons from GitHub's First Year
- The Innovators
- Particle Physics for Non-Physicists
- What You Do Is Who You Are
- The Startup of You
- How to Avoid a Climate Disaster