Wonjun Jang Email & Phone Number

# Kakao Corp. AI Model Development Team1. Large Language Model End2End Training- Include Pretraining language model (PLM), Supervised fine tuning(SFT), RLHF (Reinforcement Learning From Human Feedback)- Preprocessing dataset for PLM.- PLM multi-node training and performance evaluation- Managing human labeler to create datasets for SFT and RLHF- Research on datasets that work well for SFT and Training# Kakao Corp. Recommendation Team1. Sequential RecommendationAchievements- Build transformer based sequential recommendation models from scratch.- Build c++ based dataset builder for model to execlerate preprocessing time on large dataset- Find domains where the sequential recommendation system works well through experiments.2. Recsys Challenge 2022Achievements- I conducted an experiment to see if the sequential model works well on the session based dataset, and it works well.- In the form of [view]*N + [Purchase], data was augmented by data augmentation in the form of last view prediction in addition to purchase prediction, and significant results occurred in performance. (public leaderboard ranked 15th -> 13th.)

대한민국 서울

Search & Recommender system.Machine learningGraph Neural Network based recommender system- Neural Graph Collaborative Filtering- Graph attention network- graph embedding- Improve CTR(Click Through Rate) 14%Transformer based Information retrieval model- Encode Document and Query by transformer encoder- calculate query - document attention- Improve CTR(Click Through Rate) 7%Airflow based Machine learning pipeline- Develop data preprocessing, training, deploy pipeline based on Airflow- Distributed data preprocessing on AWS ECSMachine learning model serving to AWS- Deploy pytorch model to Sagemaker Endpoint with GPU, Elastic Inference- Separately deploy business logic and machine learning models to Sagemaker and AWS ECS fargate- Make deploy script and AWS machine learning serving infraElasticsearch Search engine- Managing synonyms dictionary base on Remember career's profile data- Develop ES Ranking score function- Develop mysql -> Elasticsearch ETL pipeline with python and Airflow

Software Developer At Peoplefund

- 채권 정산 시스템 개발 2019.6 - 2020.5P2P banking system 개발 전문통신을기반으로고객출금시스템은행과직접연동및개발 Celery 운영 issue trouble shooting2019.1 - 2020.05 Skill set: Django, MySQL, Celery, Redis- Main server dockerizing 및 AWS EC2 to ECS migration 2019.1 - 2019.12Legacy system dockerizingLegacy EC2 infra를 ECS로이관Vault를 활용한 Container Secret management system개발AWS Lambda와 Elasticsearch를이용한 Container 환경 log pipeline구축 AWS Codepipeline + Codebuild를 활용한 ECS 환경 CD구축발표자료:https://bit.ly/2rLI7MZSkill set: Docker, AWS ECS, AWS Codepipeline, AWS Codebuild, Hashicorp Vault, MySQL- Data Pipeline service 개발 2019.8 - 2019.10DW 과 google spreadsheet를 연동한 사내 data pipeline 개발Skill set: Amazon Redshift, python, Dramatiq, Rabitmq, Google apps script