Multi-modal Catalog Attribute Extraction Platform at Instacart

Instacart has developed PARSE (Product Attribute Recognition System for E-commerce), a multi-modal catalog attribute extraction platform that uses LLM (Large Language Models) to enhance accuracy and scale for millions of products with diverse attributes. PARSE supports both text and image-based attribute extraction, consolidates attribute pipelines, and includes components for configuration, ML extraction, value normalization, and quality screening. It aims to solve challenges like prolonged data acquisition and limited attribute coverage, ensuring faster and more precise attribute data generation for improved customer experience.