Using Jev for RAG as a Similarity Metric and Reranker

Melvin Vivas · X post · 2026-09-25 · Open on X

Topics: Retrieval-Augmented Generation (RAG), Embeddings & Vector Databases · Level: intermediate

Summary

The post quotes a claim that Jev's semantic matching is more reliable than plain dot-product similarity for RAG. With small datasets it can be the direct similarity metric. With large datasets it works best as a reranker after a first retrieval step.

Key points

• Plain dot-product embedding similarity can be less reliable than a dedicated semantic-matching model.
• Small data: use Jev directly as the similarity metric to score every candidate.
• Big data: first retrieve candidates cheaply with vector search, then use Jev as a reranker.
• This is the common two-stage RAG pattern: fast retrieval first, then more accurate reranking.

Resources mentioned

• Jev · tool · typesafe.ai · paid
  A decision model that makes turn-level forecasts on AI agent conversations using only structural signals (turns, tool calls, workflow stages, timing).
  Also in: Generating a Repo Promo Video with a Claude Skill on Sonnet 5.5 vs Opus 5.5 (Melvin Vivas on X · notes), GLiDE by Fastino Labs: A Post-Trainable Reasoning Decision Model (Melvin Vivas on X · notes), Using Jev as a Reranker to Augment RAG Retrieval (Melvin Vivas on X · notes), Using Jev (TypeSafe AI) as a Decision Model for Email Classification (Melvin Vivas on X · notes) and 12 more

Try this

• Try Jev as a reranker after vector retrieval in your RAG pipeline.
• For small datasets, compare Jev scores with dot-product similarity.
• Benchmark retrieval quality of dot-product similarity vs. Jev reranking on a small RAG dataset.