Building & Managing MLOps Pipelines in the Cloud
summary
Describes building cloud MLOps that automatically run parallel training jobs (example: 40 fifteen‑minute jobs run concurrently instead of sequentially), package code in Docker and deploy via SageMaker, or orchestrate on EKS + Kubeflow, with options for serverless and asynchronous inference. Covers concrete pipeline components, zero‑downtime model swaps, centralized logs, autoscaling pitfalls, data‑drift and fairness checks, and hard security controls like network isolation and infrastructure‑as‑code. Emphasizes testing and failure modes rather than a turnkey success story.
AI-written summary · about
chapters (16)
- 0:00MLOps pipelines intro
- 2:15Machine learning process
- 4:43MLOps pipeline challenges
- 6:09Kubernetes for MLOps
- 8:38Cloud for ML experiments
- 10:26ML pipeline components
- 12:09Flexibility vs stability
- 14:56Zero-downtime deployments
- 17:45Choosing MLOps tools
- 19:49Troubleshooting systems
- 21:14Team roles & security
- 22:59Kubernetes prerequisites
- 24:45Scaling MLOps pipelines
- 26:30Chaos testing failures
- 27:31ML pipeline security
- 30:41Securing ML pipelines
related talks
36mMLOps from the edge to on demand visualization
34mImplement MLOps practices with Amazon SageMaker
25mBuilding Scalable MLOps Pipelines That Actually Work
33mBuilding Resilient (ML) Pipelines for MLOps
29mTeam Teardown: MLOps in the Cloud for IoT
53mExperience with an end-to-end MLOps pipeline: Practical insights and 3D architectural perspectives
36mImplementing MLOps practices on AWS
10mFrom Model to Production: Building Scalable MLOps Pipelines