ArturoCortinovis

Factory-floor telemetry for 20 simulated motors — streamed, stored, and alerted on in real time.

• A simulator publishes motor telemetry over MQTT into FairCom Edge, which acts as both broker and database. A FastAPI service reads from it and serves a React dashboard with live status, charts, and alert history.
• Alerting uses hysteresis — one threshold to raise an alert and a separate, lower one to clear it — so an alert doesn't flap while a reading sits on the boundary.
• The FairCom client is a module-level singleton: get_client() is synchronous, init_client() is async. They're split because authentication is a network call, and the client is initialised once from a lifespan startup hook.
• The alert loop runs as a background asyncio.create_task alongside the request handlers.
• Docker Compose healthchecks gate startup order, so the API doesn't come up before the broker is ready to accept it.
• Known limitAlert counters are held in memory, so they reset whenever the service restarts.

A small five-endpoint service built from scratch as internship preparation — a learning exercise, kept here because it's the honest version of what I can write alone.

• Pydantic models validate every request and response body, so bad input fails at the boundary rather than three layers in.
• Shared setup is wired through FastAPI's Depends() injection instead of module-level globals.
• pytest covers each endpoint, including the error paths.
• Storage is in memory on purpose — the point of the exercise was the framework, not the database.

An Express + SQLite REST API with full CRUD, built and tested end to end.

• RESTful routes with correct HTTP methods and status codes, input validation, and a global error handler so failures return a consistent shape.
• Every query is parameterised — no string-built SQL anywhere in the data layer.
• 12 endpoint tests in Jest and Supertest covering both happy paths and error cases, with environment-based configuration.