Minjae Lee Email and Phone Number

▪️ Business Optimization and Analytics• Led semi-automation of priority area selection for new business initiatives (1/3 time of manual analysis)• Developed discount range recommendation for new restaurants to boost adoption rates▪️ Geospatial Time-Series Data Analysis / Modeling• Utilized data analytics to enable data-driven decision-making across the organization• Developed statistical/ML models to predict critical steps in the delivery process▪️ Demand Control Enhancement• Improved features to maintain consistent service quality during peak order volumes• Performed a national switchback A/B test to objectively evaluate system performance▪️ Work Automation / Dashboarding• Created nationwide monitoring dashboard and ML alerts, reducing review time (15+ min → 3 min)• Developed data-based automated decision-making dashboard, saving 5+ hours daily▪️ Machine Learning Operations• Developed Airflow DAG pipeline for streamlined data ETL and model train/deployment• Implemented incremental DAG for restaurant statistics (40x faster DB ingestion for 30+ tables)

▪︎ Conducted quantitative research on financial time series, implementing signal processing techniques for algorithmic trading in Cryptocurrency and U.S. markets.▪︎Developed semi-automated trade execution and monitoring systems using exchange and webhook APIs

▪︎HAPGNN: Hop-Wise Attentive PageRank-Based Graph Neural Network• Author: Minjae Lee, Seoung Bum Kim• Published in 'Journal of Information Sciences' (Impact Factor:8.322)▪︎Remaining life estimation of power plant circuit breakers• Developed a regression/classification machine learning model estimating the remaining life expectancy of a power plant circuit breaker on large time-series data (100+ GB).• Achieved 95%+ accuracy• Investigated main sensors that impact model accuracy using an explainable method(SHAP).• Skills: Time-series analysis, large data processing, explainable methods (SHAP), product management (PM)▪︎Table-to-Text NLP model of a battlefield simulation• Developed a Table-to-Text NLP model based on Huggingface T5 model with designing realistic input/output data structure plan alongside.• Accomplished 2nd year goal; ROUGE-F1 > 0.8• Skills: NLP, product management (PM)▪︎Casual conversation chatbot• Developed a casual conversation chatbot based on GPT-2 and telegram open API• Added rule-based filters before NLG module to prevent abnormal sentence generation• Skills: NLP, open API application▪︎In-house document summarization NLP model• Developed a lightweight Korean BERT document summarization model by utilizing ‘Knowledge Distillation’ technique with web-crawled Korean Wikipedia and news texts. Model parameters were decreased by 25% while maintaining model performance (77→76%). Other compression methods (i.e. quantization, mixed precision) were reviewed alongside.• Conducted domain adaptation via fine-tuning on B2B documents. Model performance on KorQUAD increased from 37.8% to 63.7% after fine-tuning.• Skills: NLP, model compression, web-crawling▪︎StarCraft II Reinforcement learning(RL) agent• Utilized the RL algorithm, A3C, in a StarCraft II environment with a reward function based on completing specific tasks (i.e. gather minerals, attack opponents) via PyTorch and StarCraft II simulator.• Skill: Reinforcement learning