Proposed structure: link-shortener/ ├── app/ # application code, importable as a package │ ├── main.py # FastAPI app + route wiring │ ├── db.py # SQLite connection + schema init │ ├── models.py # Pydantic request/response schemas │ └── shortener.py # short-code generation + lookup logic ├── tests/ # pytest tests against the API └── requirements.txt # pinned dependencies Each top-level folder, one phrase: app/ holds runnable application code; tests/ keeps verification separate from shippable code; requirements.txt pins what runs. Task restatement: Build a small FastAPI service backed by SQLite that lets a client submit a long URL and get back a short code, then resolve that short code via a redirect endpoint that sends the browser to the original URL. Steps: Pin dependencies (fastapi, uvicorn, pydantic) in requirements.txt — SQLite needs no package, it's in Python's stdlib. In db.py, open a SQLite connection and create a links table (code primary key, url, created_at) if it doesn't exist. In shortener.py, write short-code generation (base62 over a random or sequential id) and a lookup-by-code function. In models.py, define a CreateLinkRequest (the long URL) and a CreateLinkResponse (the short code + full short URL). In main.py, wire two routes: POST /links to create, GET /{code} to redirect. In tests/, cover create-then-redirect and the unknown-code case. Files to create: requirements.txt — declares what the project runs on. app/db.py — owns the SQLite connection and schema so other modules don't duplicate it. app/models.py — request/response shapes, validated by Pydantic at the edge. app/shortener.py — isolates code-generation logic so it's unit-testable without HTTP. app/main.py — the FastAPI app and the two endpoints. tests/test_api.py — proves create and redirect work end to end. Assumptions / open questions that could change the approach: ASSUMPTION: short codes are server-generated; you don't need custom vanity aliases. If you want user-chosen codes, POST /links needs a collision check and the code becomes optional input. ASSUMPTION: a single-file SQLite DB on local disk is fine; no concurrency or multi-process deployment concern yet. Open question: should GET /{code} return 307 (temporary) or 301 (permanent) redirect? Default below is 307 so you can change destinations later without browsers caching the old one. Security note: the create endpoint should validate that the submitted URL is actually http/https to avoid storing javascript: or file: schemes — I'll fold that into the Pydantic model. Approve this plan, amend any assumption, or ask a question and I'll write the code. Next: confirm 307 vs 301 redirect and whether codes are server-generated, then re-run with OUTPUT_MODE: code.