Build a Metrics API with NumPy and FastAPI

• ✓Explain what a NumPy array is, and read its shape, dtype and memory use
• ✓Replace a loop with a vectorised NumPy line, and measure the speed-up
• ✓Avoid the silent NumPy bugs: integer wrap-around, float precision, and views that change the original
• ✓Use axis and broadcasting, and fix a shape error by reading it
• ✓Find blanks, junk codes and wrong units in real data, and clean them without touching the original
• ✓Build a rolling-window anomaly detector, and judge it with precision and recall instead of accuracy
• ✓Serve results through a FastAPI service with path and query parameters, a request body and validated Pydantic models
• ✓Handle missing values honestly in JSON, return consistent errors, and paginate long lists
• ✓Test an API with pytest, including a parity test against NumPy and a test against a known truth
• ✓Measure an API with p50 and p95, and use the numbers to make a design decision

Build a small data service that cleans messy readings, finds the ones that look wrong, and answers questions about them over a REST API, and show with tests and measurements that it can be trusted.