Building Materials / E-CommerceJinyitao2025-07-24

MatrixOne Intelligence Powers Jinyitao's Intelligent Tile Image Search Platform

Jinyitao built an intelligent search platform on MatrixOne Intelligence, enabling image-based tile product discovery with 90% improvement in search efficiency.

Jinyitao

KITO (Guangdong KITO Ceramics Group), founded in Foshan in 2004, is a leading Chinese architectural ceramics group and was the first domestic manufacturer to produce inkjet thin-slab tiles. It runs three production bases-Foshan, Jingdezhen and Sanshui-and offers over a thousand SKUs across five product families.

90%
Search efficiency up
Real-time
Inventory queries
Multi-modal
Image + text search

The challenge

The catalogue itself is the obstacle. With thousands of tile SKUs across finishes, sizes and series, keyword and model-number search cannot keep pace with a live sales conversation: a customer describes or shows the look they want, and the salesperson has to find matching products immediately and know whether they are in stock. Product images were organized by surface type rather than by model nomenclature-so the very attribute customers care about was the one traditional search could not reach-and stock levels lived in a separate lookup.

The solution

KITO deployed an image-based product search service on OmniFabric. Product images and inventory data are ingested from the backend management systems and kept current through automatic updates and API synchronization. An EfficientNet model turns each image into a high-precision embedding vector, and retrieval combines semantic text-to-image search with image-to-image vector search. A search API returns matches with category and series filtering, surfaced in a mini-program that shows product name, specification, image and live stock level together.

The outcome

Search efficiency improved by 90%-sales teams locate matching tiles during the customer conversation rather than after it.

Image-to-image retrieval reaches a catalogue organized by surface type, where keyword and model-number search had no way in.

Results carry product name, specification, image and stock level together, removing the separate inventory lookup that used to follow every search.

Solution Architecture

Data sources
  • Product Image Library (Organized by Surface Type)
  • Product, Series & Specification Master Data
  • Inventory and Stock Levels
  • Backend Management System APIs
MatrixOne Intelligence
  • EfficientNet Feature Extraction to Image Embeddings
  • Vector Index for Image-to-Image Retrieval
  • Hybrid Semantic (Text-to-Image) + Vector Search
  • Real-Time Sync via Automatic Updates and APIs
  • Search API with Category and Series Filtering
Business applications
  • Snap-and-Find Product Search
  • Look-Alike Recommendations Across Series
  • Live Stock Check in the Same Result
  • Mini-Program Access for Field Sales

Related solutions and products