Now

Working on#

• Re-architecting a Life Cycle Assessment platform to be AI-native: retrieval over regulatory and product data, agentic workflows replacing form-heavy flows, and the eval harness that keeps both honest.
• A permission-aware RAG agent over company knowledge (Notion + Drive), with an eval set gating every pipeline change.
• Writing one post a month here about what breaks along the way.

Learning#

• Retrieval evaluation beyond hit-rate: graded relevance and answer-level metrics on French-language corpora.
• Cost and latency observability for LLM pipelines in production.

Reading#

• AI Engineering — Chip Huyen.
• Papers and posts on agent evaluation; the gap between demo and production remains the whole story.