Treat Your Index as a Fine-Tuned LM: Weaviate Podcast on RAG for Query Understanding
CShorten30 · x · 2026-09-16
In Weaviate Podcast #133, Trey Grainger proposes treating your search index as "a language model perfectly fine-tuned to your data": extract entities from a query, check what actually exists in the index and how it relates, and feed that back before the real search runs — effectively RAG for query interpretation, then RAG on the results.
Doug Turnbull highlighted the Purple example: a brand that's also a color, where one aggregation over your own data plus an LLM resolves which mattress the shopper means.
They also debated whether agentic search loops reduce the need for ever-better embedding models, and how coding agents are reshaping their work.
More from coding & agent
- Clay built agents on Vercel AI SDK; LangSmith tracing took one line — LangChain · 2026-09-17
- Sentry founder questions the value of WebMCP, asks for a killer use case — zeeg · 2026-09-17
- Sentry CEO David Cramer doubts WebMCP's value, clashing with Wes Bos's hype — zeeg · 2026-09-17
- happyrobot podcast breaks down the FDE motion and why it breaks at 60 engineers — ivory_tang · 2026-09-17
- HashiCorp founder's "whiteboard defense": can you explain any AI-assisted system you shipped? — charles_irl · 2026-09-16
- Production team has editors narrate their cuts to train future AI editing agents — alexgoughcooper · 2026-09-16