Jev vs embeddings for page selection
Sayora write-up comparing Jev with production embeddings for picking the right website page to answer caller questions.
This summary is based on how curated lists describe the project; its README was not available to review.
Overview
This article on Sayora's site describes replacing production embeddings with Jev for page selection: given a caller's question about a small business, choose which page of the business's website should be handed to the model to answer it.
According to the listing, across 168 caller questions on four small-business websites, Jev put the right page first 88% of the time versus 70% for the production embeddings, and handed the model the wrong page about a tenth as often. It is framed as a production replacement rather than a lab benchmark. The article text was not available for review, so method details such as how candidate pages were presented to Jev are not summarised here.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which page of this website best answers the caller's question?(inferred) | — |
How it works
- Caller asks a question
- Candidate site pages presented to Jev
- Jev picks the page to hand to the answering model
- Compare with embedding retrieval on right-page-first rate
Strengths
- Listing reports 88% right page first versus 70% for embeddings
- Far fewer wrong pages passed to the model
Caveats
- Small evaluation (168 questions, four sites)
- Article not reviewed in full
At a glance
- Type
- Article
- Status
- production
Use cases and integrations
How curated lists describe it
Production replacement: across 168 caller questions on four small-business websites Jev put the right page first 88% of the time against 70% for production embeddings, and handed the model the wrong page a tenth as often.
yibie/awesome-jev
Frequently asked questions
How did Jev compare with embeddings?
The listing reports the right page first 88% of the time versus 70%.
How big was the test?
168 caller questions across four small-business websites.
Is this a lab benchmark?
It is described as a production replacement.
Related projects
- jev-rerank★ 17
Jev as a calibrated reranker: query plus up to 30 documents in one call, one relevance Noul per document, benchmarked on BEIR.
ProjectbetaPython Milvus-based multi-hop graph RAG library where two-stage Jev Noul scoring selects relations and reranks passages.
LibrarybetaPythonCompares Jev's zero-shot reranking with specialist rerankers on MS MARCO passage and document benchmarks and a TREC WSJ transfer test.
ProjectresearchPythonLocal-first knowledge search with BM25 shortlisting and Jev Noul reranking by default, seven selectable pipelines and cited answers.
AppbetaPythonOpen Jev-style 0.6B decision model plugged into BlitzRank and Reranker-Guided Search, scoring k candidates per match in one pass.
ServiceresearchPythonProfile matching that encodes preference criteria as Clef or Jev decision vectors, retrieves with HNSW and checks rules both ways.
BenchmarkresearchPython