I’m building an AI-powered safety solution that watches live camera feeds for the very first signs of flame or smoke, then triggers an instant email alert so people can react before sprinklers or alarms even notice a problem. Here’s the workflow I need brought to life: • A computer-vision model—trained specifically for indoor scenes such as warehouses, offices and server rooms—runs continuously on a local server or edge device, analyses RTSP/ONVIF streams in real time, and flags a frame the moment it spots smoke or fire. • The back-end immediately dispatches an email that contains a timestamp, location tag and the annotated frame. • An Android companion app displays the same alert feed, lets a user acknowledge or dismiss incidents, and allows basic camera health checks. For practical use, latency must stay under two seconds from event to email, with detection accuracy high enough to minimise false positives. I’ll need: 1. The trained model (PyTorch or TensorFlow) plus the dataset preprocessing scripts. 2. A lightweight detection service packaged as a Docker image with REST/WebSocket endpoints. 3. Email notification module configurable for SMTP. 4. Fully functional Android app with clean, commented source. 5. Setup guide and a quick demo video proving the system works on my indoor test cameras. If you’re comfortable combining computer vision, real-time streaming, and Android development, let’s discuss your approach and timeline.