Querying and Retrieval

The "RAG Chain" logic:

def ask_gemini_with_rag(user_question):
    # 1. Retrieve
    results = collection.query(query_texts=[user_question], n_results=3)
    context_text = "\n\n".join(results['documents'][0])
    
    # 2. Augment Prompt
    prompt = f"""
    You are a helpful assistant. Answer the user question based ONLY on the context below.
    
    Context:
    {context_text}
    
    Question: {user_question}
    """
    
    # 3. Generate
    return model.generate_content(prompt).text

Retrieval Tuning

• top_k (n_results): How many chunks to retrieve? usually 3-5 is good.
• Re-ranking: For advanced apps, retrieve 50 chunks, then use a specialized "Re-ranker" model to pick the absolute best 5 to send to Gemini.

Summary

The prompt template is the glue. It forces the model to look at the retrieved text.

In the final lesson of this module, we ensure Validation.