The free RAG production checklist
A 2-page PDF with 25 checks to run before a RAG feature reaches real users. It covers the failures demos hide: missed exact-match queries, invented citations, prompt injection through documents, and quality that slips after a prompt change.
Get the PDF (free)
It is delivered through Gumroad. Enter $0 as the price and your email, and the download link arrives instantly. You'll also get new RecallRun experiments by email. Unsubscribe any time.
What's inside
- Ingestion (6 checks): heading-aware chunking, source metadata, re-ingestion without duplicates, embedding model tracking.
- Retrieval (5): hybrid search for exact tokens, latency budget, ANN recall, empty-result handling, permissions.
- Generation (6): server-side citation checks, refusals, prompt-injection hygiene, bounded context, streaming sources first.
- Evaluation (4): a golden set, separate retrieval and answer metrics, evals that fail CI.
- Operations (4): request logging, rate limits, health checks, cost per query.
Prefer to read it online? The full list is in this post, with links to the experiments behind each check.