Reranker

The Reranker class provides advanced reranking capabilities for vector search results by combining semantic relevance, vector similarity, and position-based scoring.

You initialize it with a model and a set of weights to adjust the importance of each of the three scoring components).

It has one method, rerank(), which takes a query and a set of results, and returns a set of reranked results.

If you want to have an agent running your RAG pipeline, the Reranker class can be used through the createVectorQueryTool.

Basic Usage

import { Reranker } from "@mastra/rag";
 
const reranker = new Reranker({
  semanticProvider: "agent",
  agentProvider: {
    provider: "openai",
    name: "gpt-4"
  },
  weights: {
    semantic: 0.5,
    vector: 0.3,
    position: 0.2
  }
});
 
const results = await reranker.rerank({
  query: "How do I deploy to production?",
  vectorStoreResults: searchResults,
  topK: 3
});

Reranker Initialization config

rerank() Params

rerank() Returns

The rerank method returns an array of RerankResult objects.

Here’s the type definition of a RerankResult:

Related

• createVectorQueryTool
• vectorQuerySearch