The goal of this project is to build a reliable weather prediction device for practical use in remote locations without access to accurate forecasts. The system uses machine learning to make predictions based on a combination of current weather readings and 24-hour forecasts from nearby stations. The machine learning model was trained using four previous years of weather data from three locations spatially distributed around and at a moderate to far distance from the location of interest. The model provided accurate temperature predictions with over 95% of results being within 1 degree Celsius of actual temperatures and moderate solar radiation predictions with 50% of predictions within a 25% margin of error. As a future work, the machine learning can be fully implemented on-board to increase the system's independence. Overall, the weather prediction device has demonstrated promising results and has the potential to provide reliable weather forecasts in remote locations where access to accurate data is limited. This system could be useful for those relying on off-the-grid renewable power solutions, allowing them to make informed decisions on power consumption based on accurate weather predictions.