Podcast
Mastering MLOps: Kubeflow Pipelines, Model Monitoring & Automated Retraining
Open original DataTalks.Club episode
Mastering MLOps: Kubeflow Pipelines, Model Monitoring & Automated Retraining
Original Episode
Use these links for the canonical episode and media sources.
- Open the original DataTalks.Club podcast page
- Watch on YouTube
- Listen on Spotify
- Listen on Apple Podcasts
Episode Overview
How do you build reliable, production-ready ML pipelines that detect model drift, monitor fairness, and trigger automated retraining? In this episode, Theofilos Papapanagiotou — a systems engineer with 20 years’ experience (from Unix engineering to ML engineering) now helping companies run ML workloads and a Kubeflow enthusiast — walks through practical MLOps strategies and tooling.
People
Use these links to connect the episode to guest notes.
Chapter Summary
Use these checkpoints to decide whether to open the source transcript.
- 0:00 - Podcast Introduction
- 2:34 - Episode Kickoff & Guest Overview
- 3:30 - Guest Background: From Unix Engineer to ML Engineering
- 5:14 - Defining MLOps: Culture, Process, and Technology
- 7:28 - DevOps vs MLOps: Model Lifecycle and Data Drift
- 11:17 - Monitoring for MLOps: Drift, Fairness, and Retraining Triggers
- 13:04 - Monitoring Stack: Prometheus/Grafana and Inference Sensors
- 14:44 - Commoditizing Inference Monitoring for Faster Iteration
- 15:29 - Role Distinction: ML Engineer as Practitioner, MLOps as Practice
- 16:37 - Team Composition: Developer, Operator, and Product in MLOps
- 20:08 - The “MLOps Engineer” Debate: Title vs. Cross-Functional Teams
- 23:47 - MLOps Job Signals & Maturity Models (Google and Microsoft)
- 27:01 - Maturity Levels: Manual Training → Pipeline Automation
- 30:08 - Advanced Maturity: Data-Driven Triggers and Automated Retraining
- 33:27 - Cultural Shift: Monitoring as a Source of New Training Data
- 34:25 - Tooling Landscape: Vendors, Open Source, and Kubernetes
- 37:06 - Kubeflow Ecosystem: Pipelines, KFServing, Feast, and Katib
- 40:12 - Hyperparameter Search with Katib and Notebook→Pipeline Workflows
- 42:28 - Kubeflow & TFX: ML Orchestration and Production Patterns
- 43:28 - Learning Kubeflow: Docs, Workshops, and Community Resources
- 46:01 - Getting Started: Cloud-Managed Pipelines and Simple Projects
- 46:58 - Data & Model Versioning: MLMD, Metadata, and Traceability
- 50:35 - Relationship to DataOps: Continuation and Divergence
- 51:44 - Edge & Mobile Deployment: Offline Models and Edge Kubernetes
- 54:18 - MLOps Guidance: Maturity Roadmaps and Manifesto Alternatives
- 55:13 - Why Learn Kubeflow: Community Contribution and Career Value
- 57:04 - MLOps Benefits: Automation, Productization, and Productivity
- 58:25 - AutoML & Katib: Commoditization vs. Empowering Data Scientists
- 59:49 - Simplified Serving: KFServing and Faster Model Endpoints
- 1:01:24 - Small Teams Adopting MLOps: Practical Examples and Tradeoffs
- 1:02:56 - Breaking Silos: Language-Agnostic Pipelines and Collaboration
- 1:04:59 - Closing Remarks & Resource Links