Kubeflow for enabling AI Powered Drug Discovery and development...
summary
Azimuth runs Kubeflow for end-to-end ML with concrete practices: cross-namespace PVC sharing implemented via Trident annotations and a volume reference, custom VS Code notebook images served through a reverse proxy, and self-hosted GitHub runners with horizontal autoscaling for namespaced pipeline triggers. Kyverno enforces image and workbench limits. Pipelines train 20-model ensembles on G5-class GPUs (training 6–10 days, inference hours) for an automated cough counter built from 181 + 638 day recordings and for hit selection over 1,000,000 antibodies to pick top 100. GreenOps plans use KEDA, kube-green and WASM.
AI-written summary · about
chapters (11)
- 0:00Astrogenica’s AI platform
- 1:06Kubeflow and ML infrastructure
- 3:14Storage, monitoring, CI/CD
- 5:24Cross-namespace volume sharing
- 6:04Custom IDE setups
- 6:50Self-hosted Kubeflow runners
- 7:33Security with Kyverno
- 7:55Automated cough counting
- 15:25Hit selection from data
- 21:53GreenOps initiatives
- 26:22Q&A
related talks
31mKubeflow - Data Science on Steroids
14mKubeflow: Where ML Meets Kubernetes
46mDeploying scalable Machine Learning models with Kubernetes
31mKubeflow 2026 State of the Union: Orchestrating the Full ML Lifecycle
54mScaling Machine Learning Pipelines in Cloud
14mLightning Talk: Leveraging Kubeflow for Next-Generation AI Healthcare Solutions
29mML Engineering Kubeflow
30mKubeflow, Your ML Tools on Kubernetes