End-to-End Single-Document QA

What we're building

One function, ask(), that wires together everything from Lessons 4 through 7: chunk a document, build a vector store, retrieve the most relevant chunks for a question, and generate a grounded answer. This is Naive RAG's complete four-stage pipeline, in nine lines, running against three different questions to show what it handles well and what it doesn't.

The whole pipeline, in one function

def ask(query: str, store: list[dict], k: int = 2) -> str:
    retrieved = retrieve(query, store, k)
    return generate_answer(query, retrieved)

That's genuinely the entire "Naive RAG" architecture, once chunking and embedding have already happened once to build store. Everything this course covers from here forward, better chunking, persistence, thresholds, a real vector database, is refinement around this same two-line shape, not a replacement for it.

Three questions, three outcomes

This lesson asks ask() three different questions on purpose:

1. "What's the best way to get a crispy pizza crust?" - directly answerable from one chunk, the easy case Lessons 6-7 already proved works.
2. "How does the household decide what to plant in the third garden bed?" - also answerable, but notice the chunk says the bed "rotates whatever seeds are left over," a much vaguer answer than the pizza question got. Retrieval found the right chunk; the chunk itself just doesn't contain a detailed answer.
3. "What is the capital of France?" - not in the document at all. A good outcome here is Gemini saying the context doesn't contain the answer, per Lesson 7's prompt instruction. If it answers "Paris" anyway, using outside knowledge instead of admitting the context doesn't cover it, that's a preview of a failure mode Lesson 16 covers properly: retrieval still returns its top-k chunks even when none of them are actually relevant, since "most similar" isn't the same guarantee as "similar enough to be useful."