Computer-Vision Property Condition Reports for an Investment Property Fund

Overview

Built a computer-vision system for an investment fund that inspects large batches of rental properties, automatically flagging visible damage, wear, and maintenance issues from photos taken during routine walkthroughs. Field inspectors upload photos through a mobile-friendly upload tool, and the system generates a structured condition report highlighting areas that likely need attention, cross-referenced against the property's last inspection. This replaced a fully manual review process where an analyst had to look through hundreds of photos per inspection cycle to catch the same issues. The fund's asset managers use the flagged reports to prioritize maintenance spend across the portfolio. An investment fund's asset managers relied on an analyst manually reviewing hundreds of walkthrough photos per inspection cycle to catch visible damage and maintenance issues across a large rental property portfolio. We built a computer-vision system that automatically flags likely damage and wear from walkthrough photos and compiles a structured condition report cross-referenced against each property's last inspection, so the analyst reviews flagged issues instead of every photo. We trained the detection model against a large set of the fund's own historical walkthrough photos, labeled alongside its analyst team to make sure the flagging criteria matched what they actually cared about. After validating flagged results against a full inspection cycle manually, we rolled the tool out across the whole portfolio. Manual photo-review time per inspection cycle dropped by an estimated 70%, letting the same analyst team cover a larger portfolio and giving asset managers a consistent, defensible basis for prioritizing maintenance spend.