Led the development of backend and data systems for message ad recommendations.

Managed a team of 4 software engineers, leading the development of backend and data systems for message ad recommendations and the building of ML infrastructures.- Established the mission for the newly formed team and developed a platform strategy for CURAS, achieving the 1st milestone of the platformization and resulting in various project oppotunities.- Designed and developed feature and training data pipelines, a training workflow, and a model orchestration system for the platformization project, automating the ML CI/CD/CT pipeline and enabling data scientists to develop and deploy features and models independently.- Developed leadership within the team, fostering a culture of technical leadership and agile development, resulting in a 28.6% increase in the number of successfully completed projects in 2023 despite resource constraints.

Developed backend and data systems for message ad recommendations.- Reduced the preparation time for ad targets by 97% through the design and development of an event-driven MSA system as a member of the CURAS development project, contributing to a 50% increase in message delivery for the SmartMessage service.- Designed and developed scalable backend systems for segmenting ad targets for A/B testing, caching target metadata for real-time aggregation of experimental metrics, and re-processing events idempotently to ensure testing consistency, contributing to the development of the CURAS A/B testing system.Developed backend and data systems for training ad auction bidding models.- Developed streaming fan-out and implemented Druid datasource sharding to horizontally scale the pipeline and resolve data delay and loss problems due to an increased number of ad groups. Acheived stable ingestion of over 80k RPS during peak times, supporting a 2x increase in ad groups.- Automated the activation of emergency training mode based on incoming data volume, streamliningoperations and preventing service degradation.Developed 15+ dashboards with pipelines from diverse data sources including over 5TB(1 copy) of daily logs, visualizing ad performance and anomalies within the organization.- Improved Hive job efficiency within the organization by developing and deploying an automation script for easy use of open-source JSON SerDe, reducing JSONL data processing time by over 50% compared to using Hive built-in functions.

- Designed models to predict mobile users' in-app activities.- Developed the audience targeting system for mobile advertising.- Contributed to 6.5K or more conversions including about ₩400M of purchases.

- Devised a sequential pattern mining method using genetic algorithm for in-app purchase features.- Achieved the experimental performance of AUC 83 % in identifying with the features each gamer’s intent of purchase within three days after registration.

- Proposed an ant colony optimization algorithm for the mixed Chinese postman problem.- Proved that the algorithm converges to optimality with probability 1 in the long run and its approximation ratio has the upper bound 3/2 on the optimum at each iteration.- Achieved better solutions by avg. 2.4% points in approximation ratio with the proposed algorithm than with the preceding randomized one.

- Maintained the bug/issue tracking system.- Assisted in establishing the testing criteria for the Visual Surveillance project.