Retrieval that cites its sources
A retrieval pipeline that cannot point at the exact passage it used is not trustworthy, no matter how good the answer reads.
An answer that sounds right and cites nothing is a liability dressed as a feature. The bar I’ve settled on for any retrieval pipeline I ship: every claim in the output should be traceable back to a specific chunk, and the citation should survive being clicked.
Getting there is mostly about discipline at ingestion time — stable chunk ids, source metadata that rides along through every transform — rather than anything clever at generation time. The generation step just has to be told, plainly, that an uncited claim is a failure.
Write-up on the ingestion pipeline itself coming next.
Full-stack developer and WordPress architect writing about headless CMS, Next.js and AI-assisted builds.
No spam. One or two emails a month, unsubscribe anytime.