jevbooks

← All projects

jev-retrieval

romeromarcelo/jev-retrieval · Homepage

Semantic code and document search CLI with BM25 recall and Jev calibrated precision.

Scoring & Ranking100%Runner-up: Classification & RoutingMenuFunnelTraffic light

What it is

jevr is a Rust CLI for semantic search over codebases and plain-text document corpora. It returns ranked files and line ranges with calibrated relevance probabilities, designed for developers and coding agents.

How it uses Jev

Jev Noul questions judge bounded windows of candidate files recalled by BM25. A Choice reranks survivors into grep-style path:line results. The query and selected source excerpts (up to 24 KB) are sent to TypeSafe; calibrated scores determine ranking and thresholds.

Primitives:choicenoul

Technique worth stealing

BM25 recall followed by Jev calibrated verification and reranking, with no repository state and a client-side response cache.

Try it

cargo install jevr --locked; export TYPESAFE_API_KEY=...; jevr "query" /path/to/repo

View on GitHub

judged by Jevjev-1.13.0

Evidence

Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.

  • Jev-centricyes0.89
  • Shows a System One patternunclear0.45
  • Handles uncertaintyno0.06
  • Measuredno0.06
  • Runnableyes0.94
  • Worth recommendingunclear0.54
  • Model replicano0.05
  • Problem scopescore on a 0–2 scale1.83
  • About Jevyes0.97

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 23 Sept 2026.