With the principles, infrastructure, and core platform components in place, the focus shifts to applying this knowledge in end-to-end, real-world projects. Building production ML systems involves navigating practical challenges such as managing large datasets efficiently, ensuring robust model validation, gaining deep visibility into the training process through effective tracking, and serving models reliably. This final phase integrates the concepts learned previously into tangible implementations.
This part consolidates your learning through the first two of three capstone projects: an identification card object detector and a movie recommendation system. We’ll work through data analysis and preparation using Kubeflow notebooks; build and execute training and validation pipelines by incorporating techniques such as persistent volumes for data handling and tools such as MLflow and TensorBoard for enhanced tracking; and, finally, deploy the trained models as scalable inference services using frameworks such as BentoML. These projects demonstrate how MLOps principles and tools come together to deliver value.