Mercari's Journey of User-Generated Content Translation

This article details Mercari's strategic journey in translating user-generated content for its global app, illustrating how they dramatically reduced translation costs by 100x over two years by transitioning from traditional translation APIs to Large Language Models (LLMs) like GPT and Gemini. It provides valuable insights into the technical challenges, A/B testing strategies, and non-AI features required to build a scalable and cost-effective translation pipeline for a massive consumer-to-consumer marketplace. Read it on Mercari's Engineering Blog.