AI-Native Reliability
VP of Site Reliability Engineering. I build AI-native reliability systems — instrumentation, diagnosis, and response that happen before a human is paged.
Overview — deep-dives arrive in a later phase
Incidents still wait on a human to read the alert, write the runbook, and dispatch the fix.
An agentic pipeline: a Strategist decomposes the alert, composes a dynamic runbook, and dispatches sub-agents across observability and ticketing surfaces via MCP. Human judgment stays at the decision boundary.
Pattern validated in a production enterprise environment.
independent build — available on request
Observability standards usually live in documents that no runtime enforces.
A skill library and MCP servers that encode OTel instrumentation and o11y governance — standards as running code, not review checklists — so the practice scales uniformly across hybrid-cloud platforms. Outcome quantified in the Phase 3 write-up.
standalone demonstration project
Launch-day traffic across three regions is not a load test you can rerun.
Self-service load testing validated 3M concurrent streams. 0 P0 incidents on global launch day; 100% uptime through the Olympics 2024 peak.
employer work — public metrics
SLO governance across 12 critical services and 100+ APIs.
MTTR from 1,400 hours to 60. P1/P2 incidents near zero.
employer work — public metrics
Monitoring lived as a hand-configured service-desk function, stale the day after it shipped.
95% infrastructure coverage; dashboard-as-code cut deployment time 75%.
employer work — public metrics
One channel
For a Head-of-SRE conversation, an AEGIS walkthrough, or a good argument about autonomous systems.
jeyadev.narayanan@gmail.com