Implementation:PacktPublishing LLM Engineers Handbook QueryExpansion Generate

Field	Value
Type	API Doc
Workflow	RAG_Inference
Repository	PacktPublishing/LLM-Engineers-Handbook
Source	query_expanison.py:L13-38
Implements	Principle:PacktPublishing_LLM_Engineers_Handbook_Query_Expansion

API Signature

QueryExpansion.generate(self, query: Query, expand_to_n: int = 3) -> list[Query]

Import

from llm_engineering.application.rag.query_expanison import QueryExpansion

Key Code

class QueryExpansion(RAGStep):
    @opik.track(name="QueryExpansion.generate")
    def generate(self, query: Query, expand_to_n: int = 3) -> list[Query]:
        prompt = QueryExpansionTemplate().create_template(expand_to_n=expand_to_n)
        chain = prompt | self._llm
        response = chain.invoke({"question": query.content})
        # Parses response into list of query strings
        queries = [
            Query(
                content=q,
                author_id=query.author_id,
                author_full_name=query.author_full_name,
            )
            for q in result
        ]
        queries.append(query)  # Include original
        return queries

Parameters

Parameter	Type	Default	Description
query	Query	(required)	The original user query object
expand_to_n	int	3	Total number of query variants to produce (including the original)

Inputs and Outputs

Inputs:

query: Query - The original query with content, author_id, and author_full_name fields
expand_to_n: int - The number of expanded queries to generate (default: 3)

Outputs:

list[Query] - A list of Query objects including:
- The original query (appended last)
- N-1 LLM-generated reformulations, each carrying forward the same author_id and author_full_name metadata

How It Works

A QueryExpansionTemplate generates a prompt instructing the LLM to create expand_to_n alternative phrasings of the original question
The prompt is chained with the LLM via LangChain's pipe operator
The LLM response is parsed to extract individual query strings
Each expanded query string is wrapped in a new Query object, preserving the author metadata from the original query
The original query is appended to the list to ensure it is always included in retrieval
All queries are returned for parallel vector search

External Dependencies

langchain_openai (ChatOpenAI) - LLM used for generating query expansions
opik - Observability and tracing decorator
loguru - Structured logging

Source File

llm_engineering/application/rag/query_expanison.py (lines 13-38)

Note: The filename contains a typo (query_expanison.py instead of query_expansion.py) which is preserved from the original repository.

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment