Checkpoint: Notes Search Assistant with Citations

What this is

No new concepts in this lesson. This is a checkpoint: a small, real script built entirely out of ideas from Lessons 10 through 17, combined into one thing. If you can read lesson.py and understand why every piece is there, you've mastered the Intermediate tier. If any piece feels unfamiliar, that's a sign to revisit the lesson it came from before continuing to Advanced.

What it does

Loads (or builds and persists) an embedding index over every fixture note, then answers two questions: one that needs two different sources combined with citations, and one with no relevant information at all, demonstrating retrieval's threshold correctly refusing to answer instead of guessing.

Where each piece came from

def load_or_build_store() -> list[dict]:
    if STORE_PATH.exists():
        return json.loads(STORE_PATH.read_text())
    store = build_vector_store()
    STORE_PATH.write_text(json.dumps(store))
    return store

Lesson 13 (persistence): build and save once, load on every run after that, no repeated embedding calls.

return [
    {"text": text, "embedding": vector, "source": path.name}
    for path, text, vector in zip(paths, texts, vectors)
]

Lesson 12 (metadata): every record carries its source filename.

above_threshold = [record for record in scored if record["score"] >= min_score]
return above_threshold[:k]

Lesson 14 (thresholds): retrieval can honestly return nothing when nothing clears MIN_SCORE, instead of always returning k results.

if not retrieved_chunks:
    return "I don't have any information relevant to that question."

Lesson 15 (grounded prompting): skip the API call entirely when there's nothing relevant to answer from.

context = "\n\n".join(
    f"[Source: {chunk['source']}]\n{chunk['text']}" for chunk in retrieved_chunks
)

Lesson 15 (citations): each chunk is tagged with its source before being handed to the model, so the model can cite it.

(Lessons 10, 11, 16, and 17's chunking, structure, and evaluation lessons don't appear directly in this script's code, since this checkpoint indexes whole fixture files the same way Lesson 9's did, but everything they taught about why those choices matter is what this script's design decisions rest on.)