When would I use Jev reranking over a standard reranker?

The digest frames Jev reranking for criteria too specific for a normal reranker, where each rule needs high-quality specification. Pinecone retrieves broad candidates first, then a Jev judgment narrows them with a Claude baseline alongside.

What setup turns the empty demo into working search?

Clone the repo, set the Pinecone and TypeSafe keys, run the ingest step once to populate the index, then start the dev server. Without ingest the deployed index is empty and returns nothing.

Do I need an Anthropic key just to try the reranker?

No. The Anthropic key is optional; without it the Claude comparison column stays disabled and the Pinecone plus Jev path still runs. Only the Pinecone and TypeSafe keys are required.

How do I contribute a new example without editing CI?

Add a directory under examples with its own package.json and the shared env keys. CI discovers each example and runs typecheck, lint, format, tests, and build with no workflow edit.

What does the nli-reranking example actually do?

Pinecone full-text search retrieves 200 candidates, then a TypeSafe judgment reranks them to 10 by natural-language criteria, shown alongside a Claude baseline. It suits reranking criteria too specific for a normal reranker.

What data does the ingest step load?

A fresh deploy starts with an empty Pinecone index. Running npm run ingest once loads 2,155 bird articles, and the app returns results only after that. The index and keys are your own.

What agent skills ship with the repo?

Two Claude skills: pinecone-typesafe-retrieval, covering where a TypeSafe judgment fits around a Pinecone call, including routing, picking a retrieval mode, gating a query, and reranking results; and record-demo-gif for re-recording an example's demo GIF.

CommunityJev Tools & CLIs

Using TypeSafe And Pinecone

Example collection pairing Pinecone retrieval with Jev judgments, including a reranking demo that narrows full-text candidates by natural-language criteria.

See highlights

Open github.com

Stars: 2
Language: JavaScript
Published by: Community
Author: pinecone-io
Last verified: Oct 2026

pinecone-io/using-typesafe-and-pinecone

Organization repository on GitHub

View on GitHub

Worked out examples of applying TypeSafe AI Decision (Jev) models with Pinecone.

Forks: 1
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.Homepage

About this project

A collection of examples using TypeSafe's Jev model and Pinecone together. Each one is self-contained, with its own package.json and its own deploy, and shares the API keys in the repo-root .env. More coming soon! If you have an example, or want to request one, make an Issue or PR! Use TypeSafe's Jev with Pinecone to filter out and rerank results in natural language. Great for when you have reranking criteria that aren't quite good enough for normal rerankers, and need high quality specification for each criteria.

Quoted from the project's README. Read the full README.

Tagscommunity github

Works forJev for Classification

Used inJev for Software & DevTools

Languages: JavaScript

Highlights

Each example is self-contained with its own package manifest and deploy while sharing root API keys.
Reranking demo retrieves broad Pinecone candidates, then applies Jev judgments to narrow them by stated criteria.
Ships a Claude baseline column that disables cleanly when the optional Anthropic key is absent.
Includes skills for placing Jev routing, gating, and reranking around Pinecone calls.

Watch out

Example code needing Pinecone, TypeSafe, and optionally Anthropic keys; a fresh deploy starts empty until ingest runs. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

jev-trader

Chrome side-panel bot that consults Jev each interval for trade direction on an NSE stock via the Zerodha web session, with paper trading on by default.