IoT Weather Station
Open SourceAn ESP32 weather station streaming live readings to a server that predicts local weather with two small ML models.
An ESP32 reads temperature and humidity off a DHT11 every 15 seconds and streams it over Wi-Fi. A server runs each reading through two models trained on roughly two years of local weather history — a decision tree for conditions, a linear regression for precipitation — and pushes the prediction to a live dashboard alongside the charts.
How it works
- ESP32 (Arduino/PlatformIO) → a NestJS gateway over Socket.IO → a Python subprocess holding the joblib-serialized models.
- SQLite through Prisma, mirrored to DynamoDB, with an SNS alert on out-of-range readings.
- A Next.js dashboard subscribed over WebSocket for live values and history.
The device stays deliberately dumb: the dashboard is never allowed to talk to it directly, so the sensor node carries no business logic worth getting wrong. Inference is classical rather than a neural net — two features don't need one.
A two-week university project from late 2024, built solo. Open source, with nine other single-sensor lab builds sharing the repo.