I’m developing a single platform that can forecast earthquakes, floods, and hurricanes, then turn those forecasts into timely warning alerts for end-users. To do that, the engine has to blend three incoming data streams—historical records, real-time sensor feeds, and satellite imagery—then push the results into an alerting module that can trigger SMS, email, or in-app notifications. Here’s the flow I want to see: • Data ingestion layer that routinely pulls and cleans the three sources above. • Prediction core that fuses the inputs, runs the appropriate statistical / machine-learning models for each hazard, and outputs likelihood, lead time, and confidence. • Alerting layer that converts those outputs into concise warnings with severity levels and publishes them through a REST API. I’ll provide initial sample datasets and endpoints for live sensor and satellite feeds. Your deliverable is a working prototype (code + brief documentation) that I can deploy on a cloud instance, run for a week, and see automated alerts populate a test dashboard. Please use whatever stack you’re strongest in—Python with TensorFlow, PyTorch or scikit-learn is fine—as long as it’s reproducible via Docker and comes with a short read-me explaining setup and model retraining. I’ll consider the job complete when: 1. The system ingests all three data sources without manual intervention. 2. It generates probabilistic forecasts for the selected hazards at least hourly. 3. Alerts are exposed through the API and hit my test webhook with correct severity tags. If you have experience fusing heterogeneous geospatial or sensor data for real-time prediction, that’s exactly what I need.