A walkthrough of building an AI-powered ticket triage agent using a Neo4j knowledge graph. The system models support expertise as graph relationships — who resolved which tickets, for which clients, with which skills — rather than flat attributes. A Neo4j Aura Agent reasons over subgraphs using five Cypher-based tools: semantic ticket similarity, expert ranking, client history lookup, skill matching, and SOP document search. The open-source Korca orchestration layer syncs data from Teamwork Desk, invokes the agent, and writes recommendations back. Production accuracy exceeds 90%, with data quality (clean assignment history) identified as the primary lever. The system supports manual review, auto-comment, and auto-assign modes, and is MIT-licensed for teams to fork and adapt.
