Construction Numerics – AI Construction Plan Labeling
Worked on an AI-assisted construction document intelligence project focused on labeling, validating, and improving how residential plan sets are interpreted for estimating and takeoff workflows. The project involved reviewing PDF plan sheets, classifying usable floor-plan pages versus reference-only pages, validating OCR output, identifying room labels, openings, schedules, notes, dimensions, trade scopes, and plan evidence used for estimating decisions. I also helped define and correct classification logic for architectural, framing, electrical, plumbing, HVAC, interior finish, and report-generation workflows. The work included domain-expert review of mislabeled plan content, correction of false positives, validation of page authority, and feedback on confidence scoring, exception handling, and human-review checkpoints. Quality control focused on whether the AI output matched real construction estimating logic, whether labels were useful for bid-ready takeoff reports, and whether ambiguous plan data was routed to review instead of being treated as certain.