This presentation introduces the innovative AutoNowP machine learning model for short-term weather prediction. Also known as nowcasting, short-term weather forecasts are focused on extreme weather events that represent a danger to life and property. The proposed model employs data obtained from WSR-98D meteorological radars deployed in Romania and predicts the values of future radar products such as reflectivity or wind speed using two auto encoders. This is done as a classification, meaning that, for a specific location, it is predicted if the radar value at that location will be higher or lower than a certain threshold. We introduce short-term weather forecasting focused on extreme meteorological phenomena and present the radar data set as well as the data preprocessing required for using the model. We show the innovative AutoNowP model and detail its performance when compared to other classical machine learning models.