Jev Support Desk
Developer-support tooling that routes tickets by confidence, prepares shift handoffs, and diagnoses API requests offline.
- Role
- Solo
- Stack
Problem
Support queues mix judgment calls with rules about priorities, deadlines and escalation. I wanted to measure which decisions a model can handle and send uncertain cases to a person.
What I built
One Jev request asks seven typed questions per ticket. Python handles priorities, SLAs and escalation, with a confidence gate before automatic routing.
An offline request doctor checks API request shapes, authentication headers and error responses, then produces a reproduction command without the customer’s key.
A CLI shares cached results across triage, shift handoff and evaluation commands.
Results
- On 36 synthetic labeled tickets, Jev routed 33 correctly (92%), versus 27 (75%) for keyword rules. At a 0.8 confidence threshold, 81% were auto-routed with no routing errors in that sample.
- This is a small synthetic evaluation, not a production accuracy estimate.