Sunset Learning Turns 30! Discover our story, exclusives, and celebrations →
→
- COURSE
ML Production Pipelines & Model Registry (AI PIPELINES)
Price:
Duration:
Certification:
Exam:
Continuing Education Credits:
Learning Credits:
The final 3-day lock-in that turns MLOps engineers and technical leads into owners of repeatable, auditable, enterprise-grade AI factories. Using ClearML as the single orchestration platform, you will design, execute, monitor, and govern end-to-end pipelines that automatically go from raw data → fine-tuning → evaluation → registry → deployment → retraining — with zero manual steps and full reproducibility. Thirty 30-minute labs, zero theory-only blocks. Every student leaves with a production pipeline that can retrain and redeploy their private model on new data with one click.
Upcoming Class Dates and Times
All Sunset Learning courses are guaranteed to run
Course Outline and Details
Prerequisites
- Completion of AI-301 through AI-305 (or equivalent real-world experience)
- Basic understanding of Docker & Git
Target Audience
- MLOps engineers & technical leads
- Teams who need AI that never goes stale
- Anyone responsible for governance and audit trails
Course Objectives
- 30+ fully automated ClearML pipeline runs
- Data → train → eval → registry → serve loops
- Model promotion workflows with human approval gates
- Continuous retraining triggered by drift or schedule
Course Outline
Pipeline Foundations: From Scripts to DAGs
- The Controller Pattern: How ClearML Pipelines Work
- Building DAGs: Dependencies and Data Flow
- Lab: Creating Your First Pipeline Controller
- Lab: Passing Artifacts Between Steps
- Lab: Visualizing Pipeline Execution in UI
Componentizing the Workflow
- The "Step" Architecture: Decoupling Logic
- Dynamic Parameter Injection
- Lab: Building the Data Ingestion Step
- Lab: Building the Training Step (Parametrized)
- Lab: Building the Evaluation Step
Model Registry & Governance
- The Lifecycle of a Model: Draft, Staged, Production
- Metadata Standards & Audit Trails
- Lab: Publishing a Model via Pipeline
- Lab: Querying the Registry for "Best Model"
- Lab: Locking and Archiving Legacy Models
Quality Gates & Human-in-the-Loop
- The "Safety Stop": When to Pause Automation
- Conditional Logic in Pipelines
- Lab: Implementing an Automated Quality Gate
- Lab: Adding a Human Approval Step
- Lab: Slack/Email Alerts for Approval Requests
Automation: Triggers & Schedules
- Event-Driven MLOps: Triggers vs. Cron
- Detecting Data Drift as a Trigger
- Lab: Scheduling Nightly Retraining Runs
- Lab: Triggering on Dataset Updates
- Lab: Handling Pipeline Failures & Retries
Continuous Deployment (CD) for AI
- CD Strategies: Canary, Shadow, and Blue/Green
- Closing the Loop: Updating the Inference Service
- Lab: Triggering a Model Rollout from Pipeline
- Lab: Implementing Automatic Rollback Logic
- Lab: Verifying Deployment Health Post-Update
Advanced Pipeline Patterns
- Dynamic Pipeline Generation
- Sub-Pipelines and Modular Composition
- Lab: Running Parallel Hyperparameter Tuning Steps
- Lab: Cloning & Debugging a Failed Pipeline Run
Capstone: The "Zero-Touch" AI Factory
- Capstone Brief: Automating the Lifecycle
- Lab: Build Ingest-Train-Eval Sequence
- Lab: Implement Logic for "Champion vs. Challenger"
- Lab: Configure Registry Publication & Deployment
- Lab: Simulate New Data & Verify Auto-Retrain
Continuing Education
- Continuing Education (CI/CD Integration: GitHub Actions)
- Curriculum Path: Managing Multi-Tenant Pipelines