From ML Repository to ML Production Pipeline

In this talk, Jakub Witkowski and Dariusz Adamczyk from Roche Informatics discuss their framework for transitioning machine learning models from repositories to production environments. They detail the tools and challenges involved in standardizing and scaling this process, focusing on separating ML code from MLOps code, ensuring code quality with pre-commit hooks and dev containers, and leveraging reusable components within Kubeflow pipelines. The presentation highlights how this approach enables flexibility for diverse use cases while maintaining robust standards for operationalization.