Reranking & relevance — page 2

Measured experiments on decision models. Can an open-weight model re-rank search results? Does a token budget explain a benchmark failure? What does one typed decision cost across model families? Every request and response is recorded, and negative results are published as they came out.

Experimental Dice Chess webhook bot powered by TypeSafe Jev, with adaptive legal-turn selection and no heuristic pre-ranking.

Fail-open two-hop skill suggestion plugin for DeepSeek Harness (DSH): TypeSafe Jev ranks the roster each user turn and injects one-line skill relevance hints

Internet search and page reading for coding agents, with TypeSafe Jev ranking

Local-first web app that pulls a GitHub repo's issues and README, then asks Jev (TypeSafe AI's decision model) to rate each issue by complexity, criticality, effort and relevance. Filter, dismiss, sort by weighted priority and export as plain text. Keys stay in memory; results stay in your browser.

pairsort: rank anything with AI judges — many small pairwise questions, coupled into one calibrated ranking (PKPD / Bradley–Terry) on Jev-style judges

Turn a jobs JSON file into a shortlist with TypeSafe Jev: explicit relevance criteria, source links preserved, uncertainty routed for review. Python and TypeScript.

Score AI agent context for relevance with Jev, and prune what's no longer useful.

Portable Jev skills for intent, completion, routing, Camofox browser loops and shadow context relevance.

Calibrated Jev ranking before lossless context condensation

Calibrated Jev ranking before lossless context condensation

A playground for TypeSafe AI's Jev (System One model), built around five real production workflows: support triage, RAG relevance gating, agent action firewall, inline moderation, and CI eval judging.

Drop-in rerank API served by Jev: speaks the Cohere, Jina and Voyage rerank protocols, so any RAG stack switches by changing a URL.

Jevtrieval is a retrieval-augmented generation demo. It retrieves documents from Qdrant, scores their relevance with Jev, and generates an answer from the selected documents.

Local Nevada statute search and a reproducible evaluation of Jev-assisted ranking against keyword search.

Fast, token-efficient web research for AI agents: a Bun CLI that searches the web, news, Reddit, Hacker News, GitHub, Stack Overflow, arXiv and more at once, and reads pages/PDFs. Jina Search/Reader, TypeSafe Jev ranking. Claude Code plugin.

Federated knowledge retrieval engine for AI agents. Routes queries across specialized backends (FTS5, ChromaDB, graph) via a registry-first orchestrator with TypeSafe/Jev AI routing and evidence re-ranking. Read-only by default, security-gated by design.

Real-time ad ranking and creative review on a typed evaluation model — one call prices a whole auction, no trained CTR model, no logged clicks.

Ask a private Discord channel's pile of GitHub bookmarks in plain language. Jev judges what each repository is for, how it is consumed and how ready it is, then ranks the whole catalogue per question with a relevance figure and a separate confidence figure.

Benchmarks of TypeSafe Jev on Finnish text: reranking, code search without an index, intent classification, citation checking and lecture-transcript reranking.

A local second brain built on GBrain, with TypeSafe Jev for search ranking and page connections.

Semantic command palette (Ctrl+K) ranked by intent with TypeSafe Jev — one Choice call, the probability distribution IS the ranking

Flock-camera-aware privacy navigation: see ALPR cameras near you and get driving routes ranked by camera exposure (TypeSafe Jev ranking + heuristic fallback). Paste your own key, find clean routes.

TypeSafe Jev experiments via Vercel AI Gateway: pixel B&W/gray images + long-context relevance filter bench