Technical founder. I build AI agents, MCP servers and evals that run in production, and I operate them.
- Level Up — LLM systems for businesses, from discovery to 24/7 operation: 30+ systems shipped, 8 agents running around the clock.
- Fabrik — SaaS for ISO 7010 evacuation plans, built full-stack from scratch: 400+ users, bootstrapped and profitable.
- Entrepreneur'IA — podcast with founders, researchers and operators shipping AI in production.
| Project | What it does |
|---|---|
| docs-rag-evals | RAG over the Supabase docs on Postgres + pgvector, a 70-question eval harness that gates CI, and an MCP search tool. |
| agentguard | Checkpoint, detect, recover: a reliability layer for AI agents (loops, timeouts, cost spikes, rollback). Python. |
| cache-rules | Measures your real Claude Code prompt-cache hit rate, tokens and cost from local transcripts. CLI + Claude Code plugin. |
| vocali | Dictation app for macOS, Windows and Linux: Whisper transcription, LLM cleanup, keys in the OS keychain. |
| optivision | Zero-shot object detection with Gemini: name a class in plain text, get bounding boxes back. |
Python · TypeScript · Next.js / React · Postgres & Supabase · Claude API & Claude Code · MCP · Docker
Anthropic certifications: Building with the Claude API · Claude Code 101 · Claude Code in Action · AI Capabilities and Limitations