Discover how LINE Shopping TW leverages large language models (LLMs) to extract product attributes from over 20 million items, powering key applications like auto-cataloging, combo search, and intent detection. This session shares practical strategies for building scalable, cost-efficient prompt pipelines, along with lessons learned from real-world deployment. Topics include prompt engineering, few-shot learning, and balancing accuracy with performance. Ideal for teams applying LLMs in production-scale environments.