Augmentation

Retrieval found the relevant chunks, but an LLM doesn't automatically know they exist, they need to actually be inserted into the prompt sent to the model. This step is called augmentation, and it's the "A" in RAG.

Building the augmented prompt

def build_prompt(query, retrieved_chunks):
    context = '\n\n'.join(chunk['text'] for chunk in retrieved_chunks)
    return (
        "Answer the question using ONLY the context below. "
        "If the answer isn't in the context, say you don't know.\n\n"
        f"Context:\n{context}\n\n"
        f"Question: {query}\n"
        "Answer:"
    )

# standing in for the chunks retrieve() would have found in the previous lesson
retrieved = [
    {'text': 'To reset your password, go to Settings and click Forgot Password.'},
    {'text': 'Password reset emails expire after one hour.'},
]

query = 'How do I reset my password?'
prompt = build_prompt(query, retrieved)
print(prompt)

Why "ONLY the context"

That instruction is doing real work, an LLM without it will happily blend its own training knowledge with whatever you gave it, sometimes producing a plausible-sounding but wrong answer for your specific system (a classic hallucination). Explicitly restricting it to the provided context, and telling it what to say when the answer isn't there, is what makes the final response trustworthy and traceable back to your actual data instead of the model's guesswork.

The final prompt formula

Every RAG system boils down to the same combination:

Final prompt = Query + Retrieved Context

Everything from Data Collection through Retrieval exists purely to produce a good Context, augmentation is where that context and the original question finally get combined into the single string the LLM will actually see.