STARWEST 2026 - DevOps

Modern cloud systems generate overwhelming volumes of telemetry—metrics, logs, traces, and events—yet incident response still relies on manual correlation, tribal knowledge, and brittle rule-based alerts. This work presents an approach to LLM-powered observability that augments traditional monitoring with telemetry reasoning to accelerate incident triage and root-cause analysis. Prashanthi proposes a pipeline that structures heterogeneous signals into a unified incident context, enriches them with service topology, deployment metadata, and SLO/SLA objectives, and guides engineers with...