Diagnostic Copilot

On-site diagnostics depend heavily on senior technicians. When a less experienced tech hits an unfamiliar failure, the options are calling a senior, guessing, or leaving it unresolved. And a wrong guess in the field can be a safety hazard, not just a repeat visit.

The first generation was a fixed troubleshooting script. This one is a reasoning engine: it builds a diagnostic plan for the specific case, then revises it every time the technician reports something new. The way a senior tech actually thinks.

Safety is a hard rule, not a suggestion. If a reported situation is hazardous, the assistant will not recommend the risky step. It only proposes safe actions to secure the situation first.

• Pulls the task context automatically from OfficeTrack's API, so the technician never re-types what the system already knows.
• Pre-fills the resolution back into the service form when the diagnostic ends.
• Shown live to an enterprise telecom client, on real cases. Approved after the demo.

The loop is plan → observe → revise. The model proposes; deterministic rules decide what is allowed to reach the technician. The engine, safety filter included, is covered by 200+ automated tests.

Model

Claude, via OpenRouter.

Planning loop

Plan, observe, revise: the diagnostic plan is rebuilt every time the technician reports something new.

Safety filter

A deterministic hard rule. Hazardous steps never reach the technician, only safe actions to secure the situation first.

Context

The task is pulled automatically from OfficeTrack's API. The technician never re-types what the system already knows.

Resolution

Pre-filled back into the service form when the diagnostic ends.

Test coverage

200+ automated tests on the engine, safety filter included.

The client approved the product after seeing it work on real cases in a live demo. It standardizes and speeds up field diagnostics, and reduces how often an on-site call needs a senior technician on the line.