Practical scripts that showcase query rewriting techniques for Retrieval-Augmented Generation (RAG). Each script is a minimal, runnable example.

• multi_query_retrieval.py — Multi-Query Retrieval. Generates alternative phrasings of a question, retrieves per query, deduplicates, and answers from combined context. Uses Chroma and OpenAI embeddings.
• hypothetical_document_embeddings.py — HyDE. Generates a plausible answer (a hypothetical document), embeds that, and retrieves nearest documents; then surfaces the results.
• rewrite_retrieve_read.py — Rewrite–Retrieve–Read. Rewrites the user query for retrieval, searches, then answers using the retrieved context.
• rag_fusion.py — RAG Fusion with Reciprocal Rank Fusion (RRF). Generates multiple queries, retrieves per query, fuses ranked lists via RRF, and answers from the fused context.
• step_back.py — Step-Back Prompting. Generalizes the original query to a broader version that can improve recall in downstream retrieval.
• least_to_most.py — Least-to-Most Decomposition. Breaks a complex question into simpler sub-questions to solve in order.

Notes common to the retrieval-focused scripts:

• Vector store: Chroma (no FAISS dependency required).
• Embeddings: text-embedding-3-small via langchain_openai.OpenAIEmbeddings.
• Chat model: gpt-4o-mini via langchain_openai.ChatOpenAI.
• API key is read from OPENAI_API_KEY (env or .env file). Scripts fail fast with a clear message if missing.

Prereqs:

• Python 3.11+
• An OpenAI API key in your environment (or a local .env)

1. Create a .env file with your key:

OPENAI_API_KEY=sk-...

2. Create a virtualenv and install deps:

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

3. Run a technique script:

# Multi-Query Retrieval
python multi_query_retrieval.py

# HyDE (Hypothetical Document Embeddings)
python hypothetical_document_embeddings.py

# Rewrite–Retrieve–Read
python rewrite_retrieve_read.py

# RAG Fusion (Reciprocal Rank Fusion)
python rag_fusion.py

# Step-Back Prompting
python step_back.py

# Least-to-Most Decomposition
python least_to_most.py