Computer-Vision Job-Completion Verification Tool for a Commercial Janitorial Contractor

Overview

Built a computer-vision verification tool for a commercial janitorial contractor that needed objective proof of completed work for large facility clients with strict service-level agreements. Crew members photograph key areas at the end of each shift, and the system flags obviously incomplete work — like visible trash or unclean floors — for a supervisor to review before the client sees it. This gave the contractor a way to catch quality issues internally before a client complaint came in, rather than finding out about a problem after the fact. The tool doesn't replace supervisor judgment; it simply surfaces likely problem photos so supervisors aren't reviewing hundreds of images per shift by hand. A commercial janitorial contractor serving large facility clients with strict service-level agreements had no objective way to verify completed work, often learning about quality issues only after a client complaint. We built a computer-vision verification tool where crew members photograph key areas at shift end, and the system flags likely quality issues — visible trash, unclean floors — for a supervisor to review before the client ever sees a problem. We trained the flagging model on a set of the contractor's own past photos, labeled together with supervisors to match what they actually consider a quality issue. The tool ran alongside manual supervisor review for several weeks to confirm flagging accuracy before supervisors began relying on it as the primary review queue. Client-reported quality complaints dropped as the contractor now catches an estimated majority of issues internally before the client ever sees them, giving supervisors objective, photo-backed proof of completed work for strict SLA clients.