Blog
I write about what I actually measure when building AI systems: what works, what doesn't and what it costs.
- How to measure RAG quality: numbers from practice
How to measure whether RAG over company documents answers correctly: a gold question set, MRR, refusals, and what helped and hurt in my project.
- RAG vs. fine-tuning: what your company needs
RAG gives AI knowledge from your documents, fine-tuning changes its behaviour. When to use which, how much data you need, and what my numbers showed.