WeatherBrain was developed for KTH's Project and Project Methods course. The team combined a Raspberry Pi and BME680 sensor with cloud storage and a web interface for weather measurements and experimental forecasts.
I led the machine-learning work, including collecting and preparing historical SMHI data, evaluating TensorFlow models and configuring prediction workflows. The project also involved moving sensor measurements into the cloud application.
After the course and the expiry of the project's cloud credits, the public interface was adapted to use placeholder data.
The former public demonstration used placeholder data and is currently unavailable. It was not a live forecasting service.