Free learning track
Building AI features that refuse to guess
Retrieval, chunking, relevance floors, and the refusal path — the part almost every tutorial skips and the part that decides whether the feature is trustworthy.
Who this is for: Developers shipping their first retrieval-grounded assistant.
What you will be able to do
- Chunk on semantic boundaries and keep breadcrumbs attached
- Set an empirical relevance floor and let retrieval return nothing
- Make refusal an explicitly correct outcome in the prompt
- Run a golden question set in CI where half the cases must be refused
The material
Articles in this track
Written and published. No sign-up, no email gate.
The most important feature of a RAG chatbot is the refusal
Everyone measures retrieval hit rate. Almost nobody tests whether the assistant correctly says "I don't know" — the failure that actually costs a customer.
Read itTell me what you're building.
Two minutes on the brief form gets you a real conversation: what the software has to do, what it will take, and an honest read on feasibility — whether or not we end up working together.