Replacing Black-Box Embeddings with LLM-Generated Interpretable Vectors

A developer proposed using an LLM to answer a fixed set of questions per document, turning the answers into interpretable vector dimensions instead of a 1536-dim black box. Kieran Klaassen open-sourced truffler, a Jev-powered Rails search library built on this idea.

2026-09-26 ~ 2026-09-26 · 2 related posts

• Non-black-box embeddings: use an LLM's answers to named questions as your search vectors — kieranklaassen · 2026-09-26
• Truffler: open-source Rails search replaces opaque embeddings with LLM-answered labels — kieranklaassen · 2026-09-26