Metro-MLOps Platform
A complete machine learning operations architecture created during my internship at Metropolia University Helsinki.
Project Description
This project delivers a complete, production-grade MLOps architecture deployed on Kubernetes. It is designed to automate, manage, and streamline the entire machine learning lifecycle for an MNIST digit classification model. At the heart of the system is Kubeflow Pipelines, which orchestrates each step of the workflow—from data preparation to model training and deployment—ensuring reproducibility and scalability.
MinIO acts as the central object storage solution, providing S3-compatible storage for datasets, model artifacts, and logs. MLflow is integrated for comprehensive experiment tracking and model registry, enabling seamless version control and comparison between different model iterations.
This platform is now being used as the MLOps base platform for the Metropolia AI Scalers project group, managing machine learning workflows for multiple projects in Finnish companies.
Project Gallery
