Wrapping It as a Service

Where we left off

This is a short recipe lesson, not a new-concepts lesson. naive_rag Lesson 24 already taught FastAPI's lifespan hook, app.state, and TestClient, in depth, this lesson reuses that exact pattern almost verbatim, the only thing that changed is which ingest()/ask() pair gets wired in. If any of lifespan, app.state, or TestClient feel unfamiliar, go read naive_rag Lesson 24's README first, this one won't re-teach them.

What actually changed from naive_rag Lesson 24

@asynccontextmanager
async def lifespan(app: FastAPI):
    app.state.corrective_state = ingest(NOTES_DIR)
    yield

One line: ingest() now returns Lesson 23's CorrectiveState (collection + grader) instead of a bare chromadb.Collection, and it's stored on app.state.corrective_state instead of app.state.collection. Everything else about the lifespan hook is identical.

@app.get("/ask", response_model=AskResponse)
def ask_endpoint(q: str, k: int = 2) -> AskResponse:
    return AskResponse(answer=ask(q, app.state.corrective_state, k))

Also identical in shape to naive_rag Lesson 24's one route, ask() itself does more work now (grading, filtering, rewriting), but the route calling it doesn't need to know that, it's still one function call.