All Case Studies
01AI / Computer Vision

ML Visual Similarity Search for 60,000+ Carpet Catalog

60,000+
Images searchable
Seconds
vs hours previously
2 Weeks
Training + build
PythonOpenCVCustom ML ModelReact Native

The Friction

What was broken

Jaipur Rugs maintains one of India's largest handmade carpet catalogs - over 60,000 unique designs spanning decades of production. When sales teams or designers needed to find carpets visually similar to a reference image, there was no digital system to help them.

The only process available was manual - staff had to browse through thousands of physical photographs and digital folders, relying entirely on human memory and visual recall. A single search request could consume hours of a team member's time, slowing down sales conversations, client presentations, and design decisions.

As the catalog continued to grow, the problem compounded. There was no scalable path forward with a manual approach.

The Solution

What we built

We built a mobile application with a custom computer vision pipeline at its core. Using Python and OpenCV, we trained a model specifically on carpet structural data - focusing on outer layer patterns, weave geometry, and design composition rather than just color matching.

The system was built and the model trained within 2 weeks. Users simply open the app, upload or photograph a reference carpet, and the system processes the image against the full 60,000+ catalog in real time, returning visually similar results ranked by similarity score.

The model was trained on carpet-specific features rather than general image similarity, making it significantly more accurate for this domain than off-the-shelf solutions would have been.

  • Custom OpenCV pipeline trained on carpet outer-layer structural patterns
  • React Native mobile app for iOS and Android
  • Real-time similarity search across 60,000+ image catalog
  • Results ranked by similarity score
  • Optimized for domain-specific carpet design features, not generic image matching
  • Built and deployed in 2 weeks including model training

The Outcome

What changed

What previously took hours of manual browsing now returns results in seconds. Sales teams can respond to client queries in real time during calls rather than putting clients on hold to search manually.

Design teams use the tool daily to identify production references, find similar existing designs before commissioning new ones, and accelerate their creative process.

The system eliminated an entire category of manual labor from the workflow - a task that previously consumed significant staff time daily is now instant and requires zero human effort beyond uploading a photo.

60,000+
Images searchable
Seconds
vs hours previously
2 Weeks
Training + build

Previous

Shopify Builds - Custom Liquid Theming + Headless Hydrogen

View Case Study

Next Case Study

Real-time IoT Temperature Monitoring for Yarn Dyeing Machines

View Case Study