ML at Coupang

Coupang has integrated machine learning (ML) extensively to enhance customer experiences across various platforms. The ML infrastructure supports three primary model types: Recommendation Systems, Content Understanding Models, and Forecasting Models. These models leverage large datasets to optimize personalization, product understanding, and predictive logistics. Large Language Models (LLMs) play a crucial role in improving existing ML models, enabling better customer interactions and understanding multilingual content. Coupang's LLM applications include Image & Language Understanding, Weak Label Generation, and Product Categorization. By employing fine-tuned open-source model architectures, Coupang enhances efficiency and performance while addressing the unique challenges associated with languages like Korean and Mandarin. The development lifecycle is characterized by exploration, model training, and deployment, utilizing tools such as Apache Zeppelin and Polyaxon for effective ML management.