KubeCon + CloudNativeCon North America 2024

How Ekstra Bladet’s Data Science team went from a small team of ML engineers, who needed to deliver quickly without deep technical infrastructure knowledge, to a rigid and proprietary ML pipeline built from AWS components and triggered by a large and chaotic Infrastructure as Code project. This made it difficult to achieve freedom and required a lot of work to implement and debug. One of the key reasons for adopting Kubernetes for our ML team emerged when we realized that we should serve all stakeholders across the JP/Politikens Hus organization, not just Ekstra Bladet. We then chose Kubernetes as our container infrastructure, which transformed the ML team into a dynamic ML ecosystem with great freedom under responsibility.

Initially, we focused on building robust frameworks for training and deploying ML models as API services and model training. Today, our ML team operates at the forefront of innovation, where we embrace GitOps principles to streamline our machine learning platform. Through careful adoption of advanced techniques such as autoscaling, scheduling, event triggers, and dynamic service deployment, we ensure seamless integration of new ML models into our infrastructure. This evolution has allowed us to effectively meet our diverse needs, while maintaining agility and scalability in our ML operations.