reranking-patterns

# Reranking Patterns
Improve search precision by re-scoring retrieved documents with more powerful models.

## Overview

- Improving precision after initial retrieval
- When bi-encoder embeddings miss semantic nuance
- Combining multiple relevance signals
- Production RAG systems requiring high accuracy

Improve search precision by re-scoring retrieved documents with more powerful models.

## Why Rerank?

Initial retrieval (bi-encoder) prioritizes speed over accuracy:
- Bi-encoder: Embeds query and docs separately → fast but approximate
- Cross-encoder/LLM: Processes query+doc together → slow but accurate

**Solution**: Retrieve many (top-50), rerank few (top-10)

## Pattern 1: Cross-Encoder Reranking

```python
from sentence_transformers import CrossEncoder

class CrossEncoderReranker:
    def __init__(self, model_name: str = "cross-encoder/ms-marco-MiniLM-L-6-v2"):
        self.model = CrossEncoder(model_name)

    def rerank(
        self,
        query: str,
        documents: list[dict],
        top_k: int = 10,
    ) -> list[dict]:
        """Rerank documents using cross-encoder."""

        # Create query-document pairs
        pairs = [(query, doc["content"]) for doc in documents]

        # Score all pairs
        scores = self.model.predict(pairs)

        # Sort by score
        scored_docs = list(zip(documents, scores))
        scored_docs.sort(key=lambda x: x[1], reverse=True)

        # Return top-k with updated scores
        return [
            {**doc, "score": float(score)}
            for doc, score in scored_docs[:top_k]
        ]
```

## Pattern 2: LLM Reranking (Batch)

```python
from openai import AsyncOpenAI

async def llm_rerank(
    query: str,
    documents: list[dict],
    llm: AsyncOpenAI,
    top_k: int = 10,
) -> list[dict]:
    """Rerank using LLM relevance scoring."""

    # Build prompt with all candidates
    docs_text = "\n\n".join([
        f"[Doc {i+1}]\n{doc['content'][:300]}..."
        for i, doc in enumerate(documents)
    ]