Pluggable anomaly detection pipeline integrates GAM, Superset, and Prebid data sources through custom adapters feeding a Detection Engine with Baseline Calculator, Anomaly Detector, and Dedup & Correlator components. Severity Router triggers an LLM Agent (Claude Sonnet) executing a ReAct Loop that queries Superset, GitHub, Sentry, and GAM/Prebid tools to diagnose root causes before routing alerts to Teams, JIRA, and Slack. PostgreSQL stores baselines, anomalies, alerts, and a pgvector-powered knowledge base while GCS Bucket archives LLM transcripts and snapshots for audit trails. This architecture demonstrates production-grade ML observability combining statistical anomaly detection with agentic LLM reasoning for automated incident triage and response.