FastAPI

FastAPI: the framework that made Python APIs feel modern. Type hints become validation, validation becomes docs — Swagger UI for free.

▶ Watch this reel

What you'll learn

Routes & models
Dependency injection
Middleware, background & streaming
Auth & docs

Remember this

Type-hinted function signatures ARE the contract: Pydantic validates requests/responses, bad input dies as 422 before your code
Depends() gives .NET-style DI with per-request caching, teardown via yield, and test overrides — inject db/vector store/LLM clients
SSE via EventSourceResponse streams LLM tokens to browsers; middleware + BackgroundTasks + /docs OpenAPI complete the production picture

Routes & models

Decorated functions = controllers; path/query params typed in the signature.
Pydantic BaseModel: request validation + response shaping; 422s free.
async def routes run on the event loop — await LLM/vector calls directly.

Dependency injection

Depends(provider): per-request caching, chains compose, teardown via yield.
app.dependency_overrides[...] in tests — swap at the seam.
Inject db sessions, vector store, LLM clients uniformly.

Middleware / background / streaming

Middleware: request-wide concerns (request-id, timing). CORS: explicit origins.
BackgroundTasks: fire-and-forget AFTER response — never for needed results.
EventSourceResponse (SSE): re-emit httpx LLM streams to browsers.

Auth & docs

OAuth2PasswordBearer + JWT verify (RS256 for multi-service: public key verifies).
OpenAPI + Swagger UI at /docs, generated from type hints.
Run with uvicorn (ASGI). --reload = dev only.

Code: A minimal production-shaped FastAPI skeleton

from fastapi import FastAPI, Depends, HTTPException
from pydantic import BaseModel, Field
from sse_starlette.sse import EventSourceResponse

app = FastAPI(title="RAG Service")

class Ask(BaseModel):
    question: str = Field(min_length=3)
    stream: bool = True

async def get_store():
    async with VectorStore.connect() as vs:   # per-request, auto-close
        yield vs

@app.post("/ask")
async def ask(
    req: Ask,
    user: User = Depends(get_current_user),
    store: VectorStore = Depends(get_store),
):
    async def gen():
        async for token in rag.stream(req.question, store):
            yield {"data": token}
    return EventSourceResponse(gen())

# Run:  uvicorn app:app --reload
# Docs: http://localhost:8000/docs  ← Swagger UI, already complete