Demo of Kubernetes at the Edge for AI Model Deployments
summary
A full MLOps pipeline that builds an OCI container with certified runtimes, pushes it to a registry and deploys to constrained edge nodes is demonstrated working end to end: OKD and MicroShift host the runtime, the CI uses Tekton pipelines → Argo CD → OCM to distribute and manage agents, and metrics use OpenTelemetry pre-processing sent to Prometheus/Grafana. The system tolerates intermittent or air‑gapped connectivity, sends retrain triggers or labeled outliers back to the hub, and lets operators opt into slim or full stacks.
AI-written summary · about
chapters (13)
- 0:00AI at the Edge intro
- 1:27Edge Computing challenges
- 3:34AI model lifecycle loops
- 5:42Model training workflow
- 8:10Tekton inference pipelines
- 10:17GitOps model deployment
- 13:39Edge node management
- 15:12Handling connectivity issues
- 16:40OpenTelemetry monitoring
- 18:52Grafana edge metrics
- 21:02Air-gapped edge setup
- 22:05Model retraining feedback
- 24:16Q&A
related talks
45mAIoT Ops - Using Kubernetes and ML Ops to Build Edge ML Applications
10mKubernetes: Enabling Edge Workloads
03mSponsored Keynote: Redefining MLOps at Scale on Top of Kubernetes
06mEvolving with Kubernetes: Embracing Model Ops
36mEdge Computing in Kubernetes: Best Practices for the Future
05mScaling Kubernetes Deployment at the Edge with Argo and More
25mEdge-to-Core Machine Learning and Data Pipeline on k8s
10mReady, Set, Go for K8s at edge today for AI/ML with Tensorflow