Query Rewrite

Rewrite the query with an LLM, then retrieve with an existing retrieval pipeline.

Overview

Field	Value
Type	Retrieval
Algorithm	Rewrite-Retrieve
Modality	Text
Paper	Query Rewriting in Retrieval-Augmented Large Language Models

How It Works

1. Receives a query
2. Uses an LLM to rewrite the query into a retrieval-optimized search query
3. Passes the rewritten query to a configured retrieval pipeline
4. Persists the wrapper pipeline's retrieval outputs as usual

This is different from HyDE:

• Query Rewrite changes the query text before retrieval.
• HyDE generates a hypothetical answer passage, embeds it, and searches with that embedding.

Scope

This implementation covers the paper's practical inference-time rewrite-retrieve flow only. The paper's trainable/RL rewriter is out of scope for this MVP.

Configuration

_target_: autorag_research.pipelines.retrieval.query_rewrite.QueryRewritePipelineConfig
name: query_rewrite_bm25
llm: openai-gpt5-mini
retrieval_pipeline_name: bm25
prompt_template: |
  Rewrite the following question into a concise search query.
  Keep the original intent, add missing retrieval hints only when helpful, and return only the rewritten query.

  Question: {query}
  Rewritten query:
top_k: 10
batch_size: 128

Options

Option	Type	Default	Description
name	str	required	Unique pipeline instance name
llm	str	required	LLM config name (from configs/llm/)
retrieval_pipeline_name	str	required	Existing retrieval pipeline config to wrap
prompt_template	str	see below	Template with {query} placeholder
top_k	int	10	Results per query
batch_size	int	128	Queries per batch

Default prompt template:

Rewrite the following question into a concise search query.
Keep the original intent, add missing retrieval hints only when helpful, and return only the rewritten query.

Question: {query}
Rewritten query:

Usage

Python API

from langchain_openai import ChatOpenAI

from autorag_research.orm.connection import DBConnection
from autorag_research.pipelines.retrieval.bm25 import BM25RetrievalPipeline
from autorag_research.pipelines.retrieval.query_rewrite import QueryRewriteRetrievalPipeline

db = DBConnection.from_config()
session_factory = db.get_session_factory()

wrapped_retriever = BM25RetrievalPipeline(
    session_factory=session_factory,
    name="bm25",
    tokenizer="bert",
)

pipeline = QueryRewriteRetrievalPipeline(
    session_factory=session_factory,
    name="query_rewrite_bm25",
    llm=ChatOpenAI(model="gpt-5-mini"),
    retrieval_pipeline=wrapped_retriever,
)

results = await pipeline.retrieve("Who wrote the original paper on transformers?", top_k=10)

YAML / Executor

Create or reuse a wrapped retrieval config such as configs/pipelines/retrieval/bm25.yaml, then point the query rewrite config at it:

# configs/pipelines/retrieval/query_rewrite_bm25.yaml
_target_: autorag_research.pipelines.retrieval.query_rewrite.QueryRewritePipelineConfig
name: query_rewrite_bm25
llm: openai-gpt5-mini
retrieval_pipeline_name: bm25

The executor will resolve retrieval_pipeline_name, instantiate the wrapped retriever, and inject it automatically.

When to Use

Good for:

• conversational or underspecified queries
• preserving an existing retriever while improving query phrasing
• lightweight LLM-assisted retrieval baselines

Consider other methods when:

• you need the paper's trainable/RL rewriter
• you prefer dense retrieval over rewritten text search
• added LLM latency is unacceptable