🏆 Winner · Group 9
AI-Powered IT Support Ticket Auto-Triage System
- Problem
- Help-desk analysts read every ticket by hand to judge whether it is complete, what it is about, how urgent it is and who should handle it. Unclear, duplicate and high-risk tickets slow the queue, and a chatbot that answers everything makes risky requests worse.
- Design
- A fast, conservative Gatekeeper screens each ticket first: it asks for missing details, returns existing documentation for known issues, blocks prompt injection, and holds high-risk requests for a person. Only clean tickets reach a Supervisor Triage Agent, which routes them to Tech Support, Billing, Account and Escalation specialists.
- Evidence
- Ten stress tests, including a prompt injection, ransomware, a refund request, account deletion, a screenshot ticket and a multi-issue ticket. All ten pass on all three platforms, with each fix found in testing recorded.
- Why it stands out
- Success means the safest correct action: routing, asking for more information, blocking, pausing for approval or reusing documentation, not answering everything. The team measured the latency each extra handoff added and disclosed where the design would still fail.
- Built with
- Google AI Studio, CX Agent Studio and Vertex AI Agent Builder
Hover over or select the image to read about the project.