01 Core themes (CALMS, summarized)
- Culture: blameless postmortems, small batches, collaboration across roles.
- Automation: build, test, deploy, provision—repeatable pipelines.
- Lean flow: limit work in progress; visualize bottlenecks.
- Measurement: SLIs/SLOs, metrics, logs, traces—not vanity dashboards.
- Sharing: knowledge spread via runbooks and tooling—reduces bus factor.
02 Typical delivery pipeline
Figure — CI/CD trunk
Infrastructure as code (Terraform, Pulumi, etc.) keeps environments reproducible; feature flags and canaries reduce deploy risk.
03 How DevOps relates to MLOps / LLMOps
Figure — layering
You rarely choose “DevOps or MLOps”—you need both: generic delivery discipline plus ML- or LLM-specific tooling and governance.
04 Observability trio
Metrics (aggregates over time), logs (discrete events—avoid secrets in messages), and traces (request paths across services) together explain production behavior. MLOps adds model-specific signals; LLMOps adds token usage and retrieval hit rates.