Chapter 1: Query Transformation

Why Transform Queries?

User queries are often:

• Ambiguous or incomplete
• Using different vocabulary than documents
• Too short for good embedding

Query Rewriting

from langchain.chat_models import ChatOpenAI
from langchain.prompts import PromptTemplate

rewrite_prompt = PromptTemplate.from_template("""
Rewrite the following query to be more specific and searchable.
Original: {query}
Rewritten:""")

llm = ChatOpenAI(model="gpt-4")
chain = rewrite_prompt | llm

# "how does it work?" → "How does retrieval augmented generation work?"
rewritten = chain.invoke({"query": "how does it work?"})

Query Expansion

Add related terms:

expansion_prompt = PromptTemplate.from_template("""
Expand this query with related terms and synonyms:
Query: {query}
Expanded query with alternatives:""")

# "ML" → "machine learning ML deep learning artificial intelligence AI"

HyDE (Hypothetical Document Embeddings)

Generate a hypothetical answer, embed it, search with that:

def hyde_retrieval(query, llm, retriever):
    # Step 1: Generate hypothetical document
    hyde_prompt = f"""Write a short paragraph that would 
    answer this question: {query}"""
    
    hypothetical_doc = llm.invoke(hyde_prompt)
    
    # Step 2: Embed the hypothetical document
    # (embeddings will be closer to real documents)
    
    # Step 3: Retrieve using hypothetical embedding
    results = retriever.get_relevant_documents(
        hypothetical_doc.content
    )
    
    return results

Why HyDE works:

• Hypothetical answer uses document vocabulary
• Better embedding alignment with corpus
• Especially good for complex queries

Summary

Technique	When to Use	Improvement
Rewriting	Vague queries	Clarity
Expansion	Missing synonyms	Recall
HyDE	Complex questions	Precision