devwithjevJev Reranks Knowledge Search Results
You.com uses Jev through OpenRouter as a reranking layer for Knowledge in its Web Search API, selecting relevant results between retrieval and synthesis. The setup used 3x fewer tokens and achieved 84% accuracy on the Vertical RTK benchmark.
View on X cost3x fewer tokens
3x fewer tokens. 84% accuracy on the Vertical RTK benchmark.
That’s what we got using Jev through @OpenRouter as a reranking layer for Knowledge, which we recently launched in the Web Search API.
Knowledge returns structured, real-time data from authoritative sources across domains like stocks, sports, and more. For many queries, that data answers the question directly, so much of the retrieved web context is unnecessary for the synthesis model.
Jev sits between retrieval and synthesis, selecting only the most relevant results to pass downstream.
Knowledge returns structured, real-time data from authoritative sources across domains such as stocks and sports. For many queries, that data answers the question directly, making much of the retrieved web context unnecessary for the synthesis model.
Also filed under Research & data
- Simulates Text Attention
The build uses JEV to simulate the attention mechanism by approximating an attention score for each pair of words in a text. It treats texts as bidimensional entities.
- Hermes Researches YouTube, X, and LN
Nabendu Biswas uses Jev in a Hermes agent to research YouTube, X, and LN.
- Visualizes French Wikipedia Elites
The project is a data visualization of French elites on Wikipedia. Its crawl is ongoing, with new biographies arriving.
- GraphRAG with Swappable Laya and Jev Models
This project is an agentic GraphRAG pipeline with swappable local Laya or cloud Jev decision models.