← Selected work Operational AI / Incident Response
TraceCrumb First-60 Built an incident-triage application that turns live symptoms plus prior incident memory into a bounded first diagnostic branch, with explicit fallback behavior and evidence capture.
Public technical artifact — outcome measurement pending Problem
What this was built against. Incident responders lose the first minute reconstructing context and repeating old diagnostic branches when prior incident memory is scattered.
System
How the work closes the loop. Live symptom → incident fingerprint → prior-memory retrieval → first diagnostic branch + priority checks → outcome feedback → updated incident memory.
Evidence now
What an employer can safely inspect. Public repository includes Vite/React app, Supabase boundaries, AI orchestration fallback, telemetry hooks and ship tests. Recorded claim registry reports 32 static checks + 45 graph/domain tests. The repo deliberately defines first-action usefulness and time-to-resolution as proof metrics rather than claiming them prematurely. Claim boundary
What this does not prove. No evidence yet that TraceCrumb reduced MTTR or improved first-action success in a real team. Deployment readiness and local/static checks are not presented as production adoption. Next evidence event
What would upgrade the proof. Run the no-signup demo against real recurring incident patterns and log useful / partial / missed outcomes before making performance claims.
P2–P3 — tested deployable app, no MTTR result claimed · Canonical case-study page v1