I build production AI applications. Currently focused on RAG, agent workflows, and AI-native developer tools.

🚧 Postmark — a memory layer for product experimentation, currently in active development A meta-tool that sits above experimentation platforms like Statsig and Eppo (both recently acquired — Statsig by OpenAI for $1.1B, Eppo by Datadog), providing:

• Semantic search across past experiments — "have we tested anything around onboarding before?" returns ranked past experiments with AI summaries
• Pre-flight check — paste a hypothesis, get risks, similar past tests, and sample-size critique
• Lessons graph — auto-detected meta-patterns across all experiments

Built with Next.js 16, TypeScript, SQLite + sqlite-vec, Voyage AI embeddings, and Claude Opus 4.7 / Haiku 4.5. MCP server exposes search and pre-flight to Claude Desktop. ~35% complete. Public repo launches with the working demo.

🔍 gdpr-qa — RAG over the EU's GDPR regulation. Voyage AI embeddings, Qdrant vector store, asymmetric retrieval. Article-aware chunking validated against ground truth; dimension-aware collection management.

📊 claude-analytics-dashboard — AI-powered CSV analytics dashboard with Claude's tool-use API. Drop a CSV, get streaming charts, narrative insights, and chat-with-your-data. FastAPI + React + Server-Sent Events.

📄 claude-data-reporter — Command-line CSV analysis tool using Claude's tool-use API to generate statistical insights and narrative Markdown reports.

AI: Claude API · Voyage AI embeddings · MCP · RAG · agent workflows Backend: Python · FastAPI · TypeScript · Next.js · SQLite · Qdrant Frontend: React · Tailwind · streaming UIs