Multi-Agent AI Ops Platform
Production AI System for Infrastructure Observability at Geotab
40%
MTTD reduction
30%
MTTR reduction
3
Live data sources (Prometheus, BigQuery, gcloud)
Human
Approval gate before all actions
situation
Geotab's SRE team was manually triaging incidents across GKE clusters, Prometheus metrics, and BigQuery analytics - three disconnected systems with no unified signal correlation. Incidents required hopping between dashboards before a root cause could even be hypothesized.
challenge
Surface root-cause hypotheses faster across a fragmented observability stack without replacing human judgment for remediation decisions or introducing an autonomous system that could take unreviewed actions in production.
solution
Built as a Claude Code skill that runs on a schedule on my work machine during on-call weeks. It pulls my on-call schedule from Jira, sets a cron, and scans every 30 minutes for incident tickets assigned to me. The moment it detects one, an orchestrator dispatches sub-agents concurrently against Prometheus, BigQuery, and gcloud APIs through MCP servers - correlating signals and surfacing ranked root-cause hypotheses. Every proposed remediation requires explicit human approval before execution.
KEY CONTRIBUTIONS
- Pulls on-call schedule from Jira, sets a cron, and scans every 30 minutes for incident tickets assigned to me - kicks off automatically the moment one is detected
- Orchestrator dispatches sub-agents concurrently, each owning one observability domain (Prometheus, BigQuery, gcloud APIs) through MCP servers
- Human-in-the-loop approval gate prevents autonomous remediation - agents surface ranked hypotheses, humans decide on action
- Runs locally on a schedule during on-call weeks - no infrastructure to manage, zero deployment overhead
EQUIPMENT MANIFEST