Lead AI for healthcare and life sciences at NVIDIA, Paris

I'm Anthony Alcaraz, based in Paris, where I work as Lead AI for HCLS (healthcare and life sciences) at NVIDIA. I'm a data science and AI person at heart, and I like ideas that have both rigor and some feeling behind them, hence the Feynman quote on my profile.

On the building side, I co-wrote Agentic Graph RAG for O'Reilly with Sam Julien. The companion repo, agentic-graph-rag-skills, has 50 runnable skills and 8 notebooks built on one mocked AWS DevOps scenario. I also keep a Python and Rust clean-room rewrite of the Claude Code agent harness to study its architecture. My day-to-day stack is LangGraph, vLLM, Ollama, Langfuse, Hermes, Claude Code and Codex.

I'd love feedback from anyone running agent systems in production: how are you handling observability and evaluation beyond tracing? Tool recommendations and war stories are very welcome.