ML Visual Similarity Search for 60,000+ Carpet Catalog
The Friction
A manufacturing company with 60,000+ carpet designs had no way to quickly find visually similar products. Staff manually browsed thousands of photos - a process that took hours per request and slowed down sales and design decisions.
The Solution
Built a mobile app with an OpenCV-based computer vision pipeline. Trained a custom model to detect and match outer structural layers of carpet designs. Users upload a photo and the system returns visually similar results instantly from the full catalog.
The Outcome
Search time reduced from hours to seconds across a 60,000+ image catalog. Eliminated manual browsing entirely.
- 60,000+
- Images searchable
- Seconds
- vs hours previously
- 2 Weeks
- Training + build