Case Study: Etsy achieves better image captioning and fewer hallucinations with Patronus AI

Etsy uses Patronus AI to optimize multimodal image captions across hundreds of millions of listings

Etsy, a global e-commerce marketplace, faced quality issues with its AI system that auto-generates captions for product images. The autogenerated captions frequently contained errors and unexpected hallucinations, which hampered the seller listing process. To address this challenge, Etsy turned to Patronus AI and its Multimodal LLM-as-a-Judge (MLLM-as-a-Judge) product, specifically the Judge-Image evaluator.

Patronus AI's solution provided Etsy with a scalable way to evaluate and optimize its multimodal AI system. Using the Judge-Image evaluator, Etsy could detect caption hallucinations and other inaccuracies, leveraging the platform's reasoning and experimentation framework to iteratively improve their models. This enabled the Etsy AI team to reduce errors and enhance the quality of their image captioning system.

Etsy