Provide an internal AI application during business hours.
Monitor user impact rather than only infrastructure metrics.
Choose two appropriate practices.
Use success rate, percentile latency, throttling, and grounded-answer measures, and define alert owners, mitigation, and escalation.
Detailed explanation
These reflect availability and user-facing quality.
These reflect availability and user-facing quality.
A notification must lead to consistent action.
A notification must lead to consistent action.
Model and provider failures can be missed.
Model and provider failures can be missed.
Alerts alone do not reduce impact.
Alerts alone do not reduce impact.
Try it yourself
An example you can run in a temporary verification environment.
AWS公式CloudWatchとAIF-C01 Domain 3.4の生成AI運用・監視を確認する。Expected result
生成AI固有の品質・レイテンシ・制限をSLOへ含める理由を説明できる。Key points
- SLO
- p95
- Alert
Notes
- Environment: AWS公式AIF-C01試験ガイドとAWS公式ドキュメントの確認
- Command output formatting can vary slightly by distribution or tool version.
- Run the example in a temporary directory or process when possible.
Foundation review
Read the scope first
Check whether the command acts on the current shell, a new process, an existing process, or a file.
Verify the observable result
Use the supplied command and compare the output with the expected result.