Skip black-box embeddings: use LLM questions to build interpretable vectors

burny_tech · x · 2026-09-27

Kieran Klaassen shares a trick: instead of an embedding model spitting out 1,536 meaningless numbers, ask an LLM (Jev) questions about each document and use the answers as the vector — e.g. a double-billing complaint maps to [iscustomer, urgent, aboutbilling, needsreply] = [1.0, 0.9, 1.0, 1.0], then plain cosine similarity retrieves it. Delip Rao notes you can go further: treat the interpretable embedding as a target vector and learn a model to produce it, as in their 2023 stylometry paper — now easy for anything via Jev.

Related event: Developer Replaces Black-Box Embeddings with LLM Q&A, Open-Sources Rails Search Library Truffler(5 posts)→

Original post →

More from coding & agent

coding & agent channel →