Summary
Etsy faces challenges in managing its vast inventory of over 100 million unique items listed by more than 5 million sellers, particularly in efficiently capturing product attributes. The article discusses how Etsy has turned to large language models (LLMs) to transform unstructured data from listings into structured data, enhancing product search and filtering capabilities. This process includes the development of a scalable pipeline for attribute extraction, evaluations of model performance, and monitoring of the system’s health. The results show a significant increase in listings with complete attribute coverage, improving user engagement and conversion rates on the platform.