Developed a lightweight user representation model for Daum, a popular news portal with 10 million users, by leveraging topic keywords from articles to capture evolving user interests. This model served as a context input feature for bandit models, enhancing news recommendation quality.

Built personalized recommender systems for mm, an audio-only social network service like Clubhouse. I employed graph-based recommendation models, incorporating techniques like graph pruning and regularization to optimize performance.

Addressed cold-start issues for new comics on Kakao Webtoon and Piccoma by combining content-based representation learning and collaborative filtering. These recommender systems served as user-targeted marketing tools, optimizing the first conversion rate.

Created a context-aware recommender system for Kakaotalk Gift, an e-commerce feature within Kakaotalk Messenger. The system considered multiple contexts such as demographics, product popularity, and user history including clicked and purchased items, by integrating text and image embedding alongside collaborative filtering techniques.

Developed and maintained machine learning applications for shoppinghow, an eBay/Amazon-like e-commerce platform. I implemented product categorization and matching systems using Transformers and graph algorithms, applying parallel computing techniques to efficiently handle billion-scale data.