MLOps
MLOps Delivery Track
CI for models, progressive delivery, and observability hooks that keep inference changes boring for on-call.
- Duration
- 6 weeks · hybrid
- Format
- Cohort + on-call shadowing sim
- Level
- Advanced
- Tuition (informational)
- ₩1,950,000
Program narrative
We wire training artifacts into CI gates, canary deploys, and tracing spans that surface model version IDs. Labs include a staged rollback where the model registry is the source of truth.
What is included
- · Model registry policies with immutable artifacts
- · Canary analysis notebook with Bayesian hints
- · Tracing span design for batch and online inference
- · Feature flag interplay with model versions
- · Synthetic traffic probes with SLO burn alerts
- · Post-deploy review agenda template
- · Mentor review of your on-call playbook edits
Outcomes you can demo
- · Land a CI gate that blocks undeclared dependency drift
- · Author a canary decision doc with explicit abort criteria
- · Map tracing fields to ownership in your org chart
Mentor of record
Mateo Silva
SRE manager for inference fleets; teaches progressive delivery with empathy for pager load.
Participant questions
Kubernetes depth?
We expect baseline kubectl comfort. We do not teach cluster bootstrap from zero.
Cloud vendor?
Examples use AWS-flavored services; patterns map to others with instructor pointers.
What is excluded?
Hardware procurement and capacity planning for GPU fleets are referenced but not designed end-to-end.
Recent participant notes
“MLOps Delivery Track’s canary notebook is now attached to every inference change ticket.”
“Shadowing sim was intense but kind. Still processing the rollback timeline exercise.”