AI & LLMs

Applied AI in production: evaluation before deployment, RAG that survives real data, and models treated like software, not magic.

field notes

AI & LLMs

2 July 2026 · 5 min read

Evaluating LLMs before they reach production

Demos lie. A practical method for building eval sets from real tickets, gating model changes in CI, and knowing when to fine-tune.

AI & LLMs

12 June 2026 · 6 min read

RAG systems that survive contact with real data

Retrieval quality beats model size. Chunking as a product decision, citations as a guardrail, and the one metric that matters.