This session explores DSB’s AI-powered track safety app deployed at Copenhagen train stations, focusing on the detection of "person on track" incidents using computer vision and machine learning. You will learn how the team tracked and detected persons on tracks with advanced ML models, and the key insights gained from real-world deployment. The talk covers the end-to-end machine learning workflow using MLflow, including managing hyperparameters, capturing training metrics, saving model artifacts, and handling Python dependencies. Discover practical tips on using the MLflow model registry and deploying models either through serving endpoints or direct integration within applications, ensuring robust and scalable safety monitoring at busy train stations.