You-jin Lee

Data Scientist

Seoul, South Korea

About

Experience

  • Senior Data Scientist at AB180 - 에이비일팔공
    Jan 2022 - Oct 2024 · 2 yrs 10 mos

    Retrieval-Augmented Generation (RAG)-based Question-Answering System - developed a RAG pipeline providing answers about Airbridge Product using official guide documents and internal Q&A records as a knowledge base - applied prompt engineering to provide reference blocks for the LLM response Customer Lifetime Value (LTV) Prediction Algorithm - developed a long-term retention curve prediction model based on the first N-day retention data - performed benchmark for more than 120 clients, and developed a new preprocessing method to improve prediction accuracy Marketing Mix Modeling SaaS Solution - developed a non-linear regression model to measure advertising performance using daily aggregated data - developed a budget optimizer to maximize conversion metric based on the given budget - automated batch-update and metric tracking process, built artifact store using MLflow Causal Inference-based Advertising Performance Measurement Solution - applied Propensity Score Matching (PSM) to estimate the effect of each advertisement in user journey - compared various classification models, such as Logistic Regression, LightGBM, based on the metrics including classification accuracy, feature balance to decide the best model

  • Machine Learning Engineer at 식스티헤르츠 주식회사
    Feb 2021 - Jan 2022 · 1 yr

    Automated Solar Power Prediction PoC Pipeline - provided ensemble prediction by combining the Global Forecast System (GFS) and Local Data Assimilation and Prediction System (LDAPS) to optimize incentives for small-scale electricity brokerage business - automated PoC pipeline, including data preprocessing, model prediction and client integration using Airflow Virtual Power Plant in South Korea - applied market operation rules to reverse-estimate the total ESS capacity in South Korea - built a virtual power plant by aggregating all solar and wind power plants across South Korea, and applied ESS charging/discharging rules to predict power generation of a virtual plant Renewable Energy Potential Map - developed a machine learning model (Model Output Statistics) to correct the solar radiation and wind speed prediction errors of the Global Forecast System (GFS) - applied Solar Power Conversion, Wind Turbine Power Curve Fitting to simulate power generation

  • Data Scientist at (주)우아한형제들 (Woowa Bros.)
    Sep 2019 - Jan 2021 · 1 yr 5 mos

    Review Image Censoring System - developed a classification model to block the inappropriate images such as those containing sexual, personal information included - applied weakly supervised object localization to provide visual explanations of the model’s classification results Baemin Mart Recommender System‬ - developed a recommendation algorithm for suggesting related products, applied a content-based model - stored related-product recommendation lists in batch tables, and conducted A/B testing Automated A/B Testing System - developed a data cleaning pipeline to process billions of log data to aggregate the conversion metrics at each funnel stage - provided metrics such as p-value and statistical power to assess the significance of experiment results

  • Data Scientist at HAEZOOM 해줌
    Feb 2017 - Sep 2019 · 2 yrs 8 mos

    Satellite Imagery, Numerical Weather Prediction based Solar Power Forecasting Algorithm - developed an algorithm to estimate surface solar irradiance by predicting future satellite images - developed an error-correction algorithm to reduce the bias of Numerical Weather Prediction (NWP) models 3D modeling-based BIPV Solar Power Prediction Algorithm - estimated shading area on solar panels based on 3D modeling - developed an algorithm to estimate solar power loss due to shading, considering physical characteristics (e.g. module, inverter, etc.) of the solar plant Solar Power Plant Monitoring & Anomaly Detection System - developed a probabilistic deep learning model that returns the expected output and probability of a solar plant using more than 1.3 million satellite images and meteorological observation data - performed anomaly detection by comparing the model output and the actual power generation of the plant, and also provided diagnostic results