MLOps Platform for Serving Machine Learning Models

The challenge

The client needed a controlled environment to serve machine learning models and manage their lifecycle.

The approach

Kumulus structured MLOps practices for model versioning, deployment, monitoring and controlled evolution on Microsoft Azure, applying DevOps practices across two models.

The outcome

Analytical processing moved off an overloaded operational system, with controls added for model deployment and monitoring. The Azure ML and Synapse workflow serves two machine-learning models from a separate analytical copy of up to 5 TB and returns results automatically to the transactional database.