AI Copilot Development

Description

Our field engineers carry a 900-page equipment manual on a tablet and, in practice, phone the two people in the company who have memorised it. Those two are retiring within eighteen months. We want a copilot in the field app that answers procedural questions from the manuals and the service history of the specific unit the engineer is standing in front of. It must cite the manual section, and it must work with intermittent connectivity — a lot of these sites are basements and plant rooms. Offline behaviour is a real requirement, not a nice-to-have. Degrading to cached manual search with no generation is an acceptable answer; failing with a spinner is not.

• Index the equipment manuals with section-level citation, including the diagrams-only pages
• Join answers to the service history of the specific asset by serial number
• Design the offline path: cached retrieval without generation, with a clear indicator of what mode it is in
• Integrate into the existing field app rather than shipping a second app
• Capture engineer feedback per answer so wrong procedures get corrected quickly

• Copilot inside the existing field app, iOS and Android
• Manual index with section citations and asset-history join
• Offline mode with explicit state indication
• Feedback capture and a review queue for flagged answers