Weather app to show actual performance of weather models
Ghost Line feature overlays past forecasts on reality to show model bias.

Beating NWS forecasts with a 22M model on home sensor data is genuinely impressive.
Data scientists, weather enthusiasts, home automation users
NeuralProphet · GluonTS · WeatherBench
Ghost Line feature overlays past forecasts on reality to show model bias.
MCP proxy prevents context-window bloat better than naive LLM forecasting.
Multi-model consensus beats single-source forecasts, but Weather Underground and NOAA already do this.
Multi-model ensemble + ML bias correction beats single-API weather apps.
Claims 8.0% error vs Zillow's 8.4%, but hard to verify without methodology.
Dark Sky clone with custom rain models, but Carrot Weather already owns this.