WorkshopSeptember 26, 2026

RAG that knows when to shut up

Designing assistants that abstain, degrade and cite honestly, because a confident wrong answer does more damage than "I don't have that".

What we cover

Per-mode relevance floors, browse intent versus question intent, guards that keep page scaffolding out of citations, and a fallback answer engine that works with no model at all. All of it is shown on a live assistant, including its failure cases.

Conferences, meetups and internal engineering days are welcome. Invite me through the contact page.

Built from

Alex OS case study and the Alex OS evaluation report.