Project idea · ungraded

Serve Your Model with FastAPI

Put a trained scikit-learn pipeline behind a small web API, so other programs can send a JSON row and get a prediction back — with input validation, a health check and tests, as a real service would have.

You will use

Milestones

Build the smallest working version first. Ticks are saved in this browser only.

01Train and save a pipeline from the capstone or house-price project with joblib.
02Create a FastAPI app that loads the model once at start-up and exposes GET /health.
03Add POST /predict with a Pydantic model describing one input row; return the prediction and probability.
04Reject bad input (missing fields, negative values) with clear 422 errors.
05Write TestClient tests for a valid request, an invalid one and the health check.

Stretch goals

Accept a batch of rows in one request.
Log each prediction with a timestamp and add GET /stats with simple counts.
Package the service in a Dockerfile.

Data

Use the model and data from the capstone (synthetic churn data you generate) or from the house-price project. See model persistence for saving pipelines.

Run it yourself

pip install fastapi uvicorn scikit-learn joblib pandas httpx
uvicorn app:app --reload

Then open http://127.0.0.1:8000/docs to try the endpoints. See running a server manually.