User repository on GitHub
Jev-powered relevance filtering and reranking for RAG in Python.
- Forks
- 2
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Repos & SDKs
Jev Reranker — hotchpotch
A Python library that reranks search results and filters retrieved documents with Jev, scoring each candidate's usefulness as evidence and dropping those below a configurable threshold before RAG context is assembled.
- Stars
- 46
- Language
- Python
- License
- MIT
- Reading time
- 4 min
- Published by
- Community
- Author
- hotchpotch
- Last verified
- Sep 2026
About this project
jev-reranker is a Python library for reranking search results and filtering retrieved documents with TypeSafe.AI's Jev. It provides prompts for both tasks and handles concurrent requests, splitting long candidate lists, and retries. For a walkthrough with examples, read Introducing jev-reranker: Reranking and Relevance Filtering for RAG.
Quoted from the project's README. Read the full README.
Highlights
- relevance_rerank() scores documents by their contribution to an answer, sorts them, and removes those below a threshold such as the default 0.2.
- rerank() reorders candidates without filtering by default; both modes take a query and a list of documents and score through the Jev API.
- Supports listwise, pointwise, and pairwise modes, and split budgets can count characters, tokens via an optional tokenizer, or a custom length function.
- Sync and async methods share concurrency limits, exponential backoff with jitter that respects Retry-After up to 60 seconds, and automatic HTTP cleanup.
- Each result carries its original document_index, and detail=True records every score including excluded documents, prompts, model, and usage.
Quickstart
python
from jev_reranker import JevReranker
query = "How long do I have to return an online order to ACME Shop?"
documents = [
"ACME Shop accepts online returns within 30 days of delivery.",
"ACME Shop in-store purchases can be returned within 14 days of purchase.",
]
reranker = JevReranker(api_key="YOUR_API_KEY")
evidence = reranker.relevance_rerank(query, documents, threshold=0.2)
for item in evidence["results"]:
print(f"{item['score']:.2f} {item['text']}")Watch out
MIT-licensed and needs Python 3.11+ plus a TypeSafe API key; Jev usage is billed, documents are sent to the API for scoring, and the optional tokenizer extra pulls in PyTorch and Sentence Transformers.
Questions about this entry
Short answers drawn from the linked project or page.
An API hop. Scoring runs through the Jev API, so every ranking call is a billed request and your document text leaves the process. In exchange the base package needs no local model or GPU, and Python 3.11+ plus a TypeSafe key is the whole requirement. Sync and async methods share the same request handling, concurrency limits, retries with jitter that respect Retry-After up to 60 seconds, and automatic HTTP client cleanup.
Relative order suffers. The relevance prompt explicitly asks for very low scores when a document only overlaps the topic, which is what makes a 0.2 cutoff useful, but pushing candidates toward zero makes their ordering less informative. So relevance_rerank() is the wrong tool when sorting the entire candidate set matters most, and rerank() is the wrong one when deciding what to keep matters most. They are different prompts, not one method with a different cutoff.
Text only. The library accepts a query string and a sequence of document strings and explicitly does not take image payloads. Long candidate lists are split automatically against a budget you can measure in characters, tokens through an optional tokenizer, or a custom length function. Returned text is untruncated, duplicate texts stay separate candidates, and ties keep input order.
No. top_k limits output, not work: the threshold is applied first and top_k does not reduce scoring work. Filtering also happens after scoring, so relevance_rerank() reduces the context you send to the next model rather than the work Jev does. The exception is degenerate input: an empty document list or top_k=0 makes no requests at all.
They are plain Python dictionaries. Pass instruction with your own instructions and criteria to override the prompt, and read the shipped ones in src/jev_reranker/instructions.py as a starting point. The README's advice is to test the prompt choice and the threshold against your own data, since lower scores and better filtering are the intent of the instructions rather than a guarantee for every document.
An empty results list, which the README frames as a signal to try another search or abstain before generating. With detail=True you still get a top-level detail record: the excluded documents with their scores and a passes_threshold flag, plus selection counts before and after, so you can see what the cutoff removed rather than only what survived.
An explicit api_key argument wins, then TYPESAFE_API_KEY in the environment, then a .env file; dotenv_path=None turns file loading off entirely. Length defaults to Python's len(), so no tokenizer or model download is required, and long candidate lists are split automatically against a budget you can measure in characters, tokens through an optional tokenizer, or a custom length function.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Search and tagging on keep.md
X: Jev reranks station suggestions in TrainLCD
YouTube: Jev in the reranking step of a RAG pipeline: why cosine similarity fails and how a steerable reranker compares. A Colab notebook comes with it.
X: LinkedIn bookmarks search
Other listings of this project
Forks and copies of this project in the directory. This entry is the canonical one.
A CLI that reranks, filters, or extractively compresses JSON retrieval results by asking Jev separate Noul questions, then lets Rust apply the order and thresholds, reading stdin and writing stdout.
Jev Tools & CLIs#community#rust#cli
Community
shinpr/jev-rerankerRust
shinpr/jev-rerankerRustNode.js CLI that pipes a JSON array of retrieved documents through Jev to rerank, filter, or compress them before they reach an LLM. Supports text and context field selection, a top-N cap, and pass-through of IDs, sources, and retrieval scores.
Jev Tools & CLIs#cli#javascript#routing
Community
More Jev resources like this
A Python library that adds .jev accessors to pandas and Polars: ask a natural-language question per row and get labels, scores, and full probability distributions back.
Jev Repos & SDKs#community#python#data
Community
ktaletsk/jevframePython
ktaletsk/jevframePythonUnofficial LlamaIndex reranker and router built on the official Python SDK: JevRerank scores retrieved passages and JevSingleSelector chooses which tool handles a query.
Jev Repos & SDKs#community#python#sdk
Community
WiktorB2004/llama-index-jevPython
WiktorB2004/llama-index-jevPythonAn open 0.6B System One decision model that reads unstructured state and typed questions in one forward pass, returning calibrated distributions with zero decoded tokens; weights are published on Hugging Face.
Jev Repos & SDKs#community#python#open-models
Community
malevrigns/agent-jevPython
malevrigns/agent-jevPythonA Python library that turns Jev-style decision models into ordinary function calls: check returns a bool, classify picks one option, label picks every option that applies, and score rates on a scale, with probabilities behind every answer.
Jev Repos & SDKs#community#github#python
Community
jlowin/vibecheckPython
jlowin/vibecheckPythonTraining code for SearchJev, a fast calibrated System-1 model for search agents: given a search state and a decision schema, it returns typed predictions over legal field values in one forward pass of a Qwen3.5 backbone.
Jev Repos & SDKs#community#github#python
Community
EvoScientist/SearchJevPython
EvoScientist/SearchJevPythonA sort CLI where the key is a plain-English description: Jev compares texts in pairs, jsort fits a Bradley-Terry scale to the answers, and prints each score with a standard error.
Jev Tools & CLIs#community#python#cli
Community
keltokhy/jsortPython
keltokhy/jsortPythonFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.