Query Rewriting Pipeline

Samira El-Masri

@samira-el-masri

·December 31, 2025

Build a multi-stage query rewriting pipeline with expansion, decomposition, and HyDE techniques to significantly improve RAG retrieval quality.

10 copies0 forks

Share this prompt:

Implement a query rewriting pipeline to improve RAG retrieval.

## Current Query Examples
{{query_examples}}

## Retrieval Challenges
{{retrieval_challenges}}

## Available Context
{{available_context}}

Build a multi-stage rewriting pipeline:

```python
class QueryRewriter:
    def rewrite(self, query: str, context: dict) -> List[str]:
        """
        Stage 1: Query expansion (synonyms, related terms)
        Stage 2: Query decomposition (complex -> simple)
        Stage 3: Hypothetical document generation (HyDE)
        Stage 4: Query variants for ensemble retrieval
        """
        pass
```

Include:
- LLM-based vs rule-based rewriting
- Caching rewritten queries
- Evaluation metrics for rewriting quality
- A/B testing framework

Details

Category

Coding

Use Cases

Query rewritingRetrieval improvementRAG optimization

Works Best With

claude-sonnet-4-20250514gpt-4o

Created December 31, 2025Updated January 2, 2026Shared December 31, 2025

Related Prompts

Evaluation Pipeline Designer

by @priya-ramanathan

Design end-to-end evaluation pipelines for model testing.

Pipeline Review Summary Generator

by @marcus-lee

Structure pipeline review summaries for leadership with risk analysis and support requests.

Data Pipeline Setup Steps

by @priya-ramanathan

Decompose evaluation data pipeline setup into steps.

Context Relevance Scorer

by @livelybear1320

Build a context relevance scorer combining similarity, keyword, entity, and topic signals to filter retrieved documents before LLM generation.

Context Relevance Scorer

by @levi-smith

Build a context relevance scorer combining similarity, keyword, entity, and topic signals to filter retrieved documents before LLM generation.

Context Relevance Scorer

by @crisdux

Build a context relevance scorer combining similarity, keyword, entity, and topic signals to filter retrieved documents before LLM generation.

More from @samira-el-masri

Context Relevance Scorer

Zero-Shot Code Bug Detection

LLM Observability Stack Setup

Negative Sampling Strategy

Create your own prompt vault and start sharing