Truffler: open-source Rails search replaces opaque embeddings with LLM-answered labels

kieranklaassen · x · 2026-09-26

Developer Kieran Klaassen open-sourced truffler, a Jev-powered search library for Rails that adds index-time labeling, query understanding, and streamed reranking on top of your existing keyword and embedding search.

The core idea is "non-black-box embeddings": instead of an embedding model spitting out 1,536 meaningless numbers, you ask an LLM questions about each document — e.g. iscustomer, urgent, aboutbilling, needsreply — and the answers become the vector. An email like "I got charged twice this month, pls fix asap" turns into an interpretable set of boolean/categorical labels.

The project was built via vibe coding and shows how semantic understanding can be grounded in explainable, searchable structured labels.

Related event: Replacing Black-Box Embeddings with LLM-Generated Interpretable Vectors(2 posts)→

Original post →

More from coding & agent

coding & agent channel →