jev-rerank-bench

Production reranker benchmark: TypeSafe Jev vs Qwen3-Reranker-0.6B on BEIR SciFact and NFCorpus — nDCG@10, cost/query and p50/p95 latency. Quality is indistinguishable; Jev is ~2x faster, Qwen ~4x cheaper. From the team at denser.ai

Does a TypeSafe Jev rerank beat embedding search? Graded relevance eval (9,831 pairs, 164 zh/en queries) over the Agent Skills Hub catalog, with the judge-circularity bias measured.

Independent Jev 1.13.0 behavior study: report, controlled prompt experiments, raw results, and offline verification.

Use Jev (TypeSafe's System One model) as a calibrated reranker: one call, up to 30 documents, a probability per document. Apache-2.0.

Windows 微信回复助手：DeepSeek 官方生成话术，TypeSafe JEV 官方判断排序，支持可取消的 3 秒自动发送。

Rerank, filter, and compress JSON search results with TypeSafe AI's Jev.

微信消息意图识别悬浮窗（macOS）：看屏 + 本地小模型判断意图和风险，再按话术生成回复候选。纯只读、不注入微信。

Minimal agent loop where Jev directs control flow and a LangChain chat model writes argument values and the final response.

Windows 微信回复助手（非官方修改版）：本地离线 OCR 读屏 + 模型起草候选回复 + 一键填入微信输入框，可选自动发送。不 hook、不注入、不读微信数据库。