I’m building a camera-based system that runs on an NVIDIA Jetson and, in real time, detects faces and recognises emotions. The entire solution must be coded in Python. For face localisation I’d like a fast deep-learning detector—SSD or YOLO—so the frame rate stays smooth on Jetson hardware. Once a face is found, a TensorFlow model should assign an emotion label (happy, sad, angry, surprised, neutral, etc.) together with its confidence score. The video stream has to overlay these results live, log every reading with a timestamp, and trigger a visual or audible alert whenever negative emotions are detected repeatedly within a short window. A lightweight dashboard served with either Streamlit or Flask will let me: • watch the annotated video feed • view rolling emotion statistics and charts • review and download the timestamped log of events and alerts Optimisation for Jetson (CUDA, cuDNN, TensorRT where appropriate) is essential, and the finished app should launch from a single command, open the dashboard in a browser, sustain real-time performance, and shut down cleanly. Please keep the code modular and well commented so I can retrain or swap models later and, if convenient, provide a Dockerfile or setup script to simplify installation.