AI Visual Search Material Library for Ralph Lauren
Challenge
Ralph Lauren's R&D teams were losing more time finding materials than designing with them. Thousands of trims, buckles, zippers, buttons, and fabrics sat trapped in static board images, with 150 or more items packed into a single image and no metadata or search behind them. Teams spent two to three hours manually browsing to locate a single item, and because designers could not easily see what already existed, materials were frequently duplicated across regions. The result was significant lost development time and avoidable duplication at scale.
Ralph Lauren's R&D teams were losing more time finding materials than designing with them. Thousands of trims, buckles, zippers, buttons, and fabrics sat trapped in static board images, with 150 or more items packed into a single image and no metadata or search behind them. Teams spent two to three hours manually browsing to locate a single item, and because designers could not easily see what already existed, materials were frequently duplicated across regions. The result was significant lost development time and avoidable duplication at scale.
Solution
Kreeda Labs built a multi-layer AI material discovery engine that turns static boards into structured, searchable data. YOLOv8 computer vision detects and isolates every item from a board image at 95 percent accuracy, OpenAI CLIP generates visual embeddings that capture colour, finish, and texture, and Tesseract OCR extracts product IDs and links them to the right items. A comprehensive metadata layer ties each material to its specifications, and an OpenSearch vector database with similarity indexing returns results in under three seconds. Designers can search by uploading an image, typing a description, or filtering by attribute. It was built with React, Node.js, and a scalable AWS deployment.
Kreeda Labs built a multi-layer AI material discovery engine that turns static boards into structured, searchable data. YOLOv8 computer vision detects and isolates every item from a board image at 95 percent accuracy, OpenAI CLIP generates visual embeddings that capture colour, finish, and texture, and Tesseract OCR extracts product IDs and links them to the right items. A comprehensive metadata layer ties each material to its specifications, and an OpenSearch vector database with similarity indexing returns results in under three seconds. Designers can search by uploading an image, typing a description, or filtering by attribute. It was built with React, Node.js, and a scalable AWS deployment.
Results
The platform cut material discovery time by 85 percent, taking searches that previously needed hours down to under three seconds. Item detection reached 95 percent accuracy, duplicate material development dropped significantly, and teams reclaimed around 40 percent of the time previously lost to manual searching, redirecting it to design work. User satisfaction was high in testing, and the system was architected to scale from an initial pilot to more than 100,000 materials globally.
The platform cut material discovery time by 85 percent, taking searches that previously needed hours down to under three seconds. Item detection reached 95 percent accuracy, duplicate material development dropped significantly, and teams reclaimed around 40 percent of the time previously lost to manual searching, redirecting it to design work. User satisfaction was high in testing, and the system was architected to scale from an initial pilot to more than 100,000 materials globally.
