Building on-Premises MLOps for ISS Columbus Ground Operations - Samo Turk

This talk explores the collaboration between Airbus and two German universities in developing anomaly detection, root cause analysis, and reconfiguration suggestion algorithms for the International Space Station's Columbus module. The speakers discuss the challenges posed by regulatory restrictions on sensor data streams, which necessitated the creation of a custom MLOps platform deployed on-premises. They share insights on how their team of AI engineers and data scientists, initially unfamiliar with Kubernetes, evolved into proficient cloud engineers by building an automated system using GitLab, microk8s, and Kubeflow. The session addresses practical considerations for bootstrapping and automating platform deployment, supporting multiple Linux distributions, managing storage, exposing services on Kubernetes, and handling identity and access management.