Building search-based RAG with Datasette

by Simon Willison @simonw

Read the original →

A real prototype log follows full-text retrieval through implementation, model changes and missing evals.

Willison records a live build that turns a question into search keywords, queries SQLite through Datasette and uses retrieved posts to answer. The account includes the annotated implementation and a model switch after keyword instructions failed. It explicitly stops short of production readiness because evaluation is missing; the dated APIs and hosted services need rechecking before reuse.

Level
Intermediate
Models
any
Category
Build logs
Tags
build-log
Source date
2024-06-21
Added
2026-10-18
Last verified
2026-10-18
Status
ok
Prerequisites
Following the annotated implementation requires SQL and JavaScript or TypeScript plus model API familiarity; no new retrieval research is assumed.
Subcategory
external builds
Region
West
Original language
en
Access
free
Resource reuse
closed: No general open-reuse grant established for this article. Free reading access does not grant redistribution rights. Facet definitions

Summary in our own words. Rights remain with the original creator. Credit and correction policy.

Plain-text Markdown copy