Checkpoint: Hybrid Search CLI

What this is

No new concepts in this lesson. This is a checkpoint: a small, real script built entirely out of ideas from Lessons 1 through 8, combined into one thing. If you can read lesson.py and understand why every piece is there, you've mastered the Beginner tier. If any piece feels unfamiliar, that's a sign to revisit the lesson it came from before continuing to Intermediate.

What it does

Builds two independent indexes over this course's six fixture notes, one dense (embeddings, Lessons 1-2), one sparse (BM25, Lessons 3-5), then answers four questions by fusing both rankings with RRF (Lesson 8) before generating an answer. Two of the questions are the bare-ID kind sparse is built for, two are the paraphrased kind dense is built for, deliberately chosen so no single retriever alone could answer all four.

Where each piece came from

def build_store(notes_dir: Path) -> dict:
    return {"names": ..., "texts": ..., "token_lists": ..., "doc_vectors": ...}

One store, both indexes: tokenized text for BM25 (Lesson 3), embeddings for cosine similarity (Lesson 2), built once, queried many times.

def hybrid_retrieve(query: str, store: dict, k: int = 2) -> list[str]:
    dense = dense_ranking(query_vector, store)
    sparse = sparse_ranking(query, store)
    fused = reciprocal_rank_fusion([dense, sparse])
    return fused[:k]

Lesson 8's fusion, unchanged: rank both ways, fuse by rank alone, no weight to tune, no scores to normalize.

def ask(query: str, store: dict, k: int = 2) -> str:
    top_names = hybrid_retrieve(query, store, k)
    return generate(query, top_names, store)

The whole pipeline as one function call, same shape as naive_rag Lesson 9's ask(), just with a hybrid retriever underneath instead of a dense-only one.