Capstone: AI SRE Assistant

Build a reliability assistant that explains service health, evaluates error-budget burn, and proposes capacity or remediation work without changing production directly.

System scope

• service catalog, SLO, deployment, and ownership context;
• metrics, logs, traces, incidents, and change-event tools;
• burn-rate and capacity analysis through deterministic calculations;
• model-generated summaries and prioritized hypotheses;
• review workflow for recommendations; and
• privacy-safe operational memory and audit.

Loading diagram...

Required evidence

Evaluate alert interpretation, calculation correctness, evidence coverage, unsupported conclusions, prioritization, latency, and operator acceptance. Compare the assistant with existing dashboards and runbooks.

Failure tests

• an SLO definition is missing or stale;
• telemetry sources disagree;
• a dashboard label contains malicious text;
• the assistant recommends capacity without cost context; and
• a user requests action outside the assistant's read-only scope.

Completion criteria

The assistant passes when every conclusion links to current operational evidence, deterministic calculations remain reproducible, and recommendations cannot bypass change control.