AI-Driven Incident Response Platform
- Designed a distributed, Kubernetes-native incident triage system with FastAPI backend services and LLM orchestration that automatically correlates alert history, recent deployments, and infrastructure state the moment an incident fires — work an on-call engineer would otherwise spend the first 10 minutes doing by hand.
- Built human-in-the-loop approval gates as a deliberate production-safety design, not a limitation: automation handles triage and context assembly and surfaces a recommended action, but a human confirms before anything changes in production.
- The engineering tradeoff was explicit — automation speed for diagnosis, human judgment for action, because an automated system that fixes the wrong thing at 3am causes more damage than the original incident.