Service Desk Operations Copilot

The problem

A lean Service Desk can lose the signal inside a loud queue. Raw ticket counts do not reliably surface a user who is blocked from working, a security item that has not been acknowledged, or a new hire whose hardware request never made it through the process.

The solution

To produce this for GitHub, I built it as a public project using entirely synthetic records. The copilot brings queue triage, onboarding correlation, a grounded analyst view, scheduled readiness checks, and an approval-gated action workflow into one interface. It is deliberately designed so the assistant can identify, explain, and draft an action, while a human remains responsible for approving any write.

Impact

• Makes work-stoppage, security, and onboarding risks visible ahead of ordinary queue volume
• Demonstrates a clear review boundary before ticket assignment, work-note, or closure actions
• Models four practical operations workflows: triage, new-hire watch, Monday readiness, and stale-ticket hygiene
• Publishes the architecture and guardrails without exposing employer data, credentials, prompts, or integrations

How it works

Lessons learned

• The better question is not whether AI can update a ticket; it is whether it can make the right risk visible early enough for a human to act.
• Least privilege and approval gates are product features. They make an operations tool easier to trust, govern, and recover.
• A credible public portfolio project needs an honest boundary: the design can be demonstrated without copying an employer system or publishing its data.