A tour of the MLOps setup at one of Denmark’s largest pension funds, PensionDenmark. I will go through how we solved a series of challenges concerning versioning, testing, monitoring, and deployment of code and ML models. Our solutions are implemented with Azure DevOps – but the principles apply to other DevOps platforms.
- How to version code and models.
- How to automatically train models with the correct code dependencies.
- How to score ML models before and during production.
- How to test deployments before and after deployment.