About

Projects Experience

• Explainable Stock Index Forecasting with Social Sentiment (Kyobo Life Insurance)

  Feb, 2024 - Dec, 202410 months

  Developed Transformer-based model encoding social media data to predict stock indices with attention-based explainability

  • Published at ICAIF 2024

  • Created interpretable AI showing which indicators influenced daily predictions

  • Integrated online social sentiment data for macroeconomic forecasting

• Text-to-Outfit Fashion Retrieval System (Fashionade & LG Display)

  Jan, 2023 - Jan, 20241 year

  Developed semantic-level fashion retrieval system integrated with Alexa voice assistant for next-gen display experience zones

  • Published at WACV 2024

  • Designed GNN-based fashion recommendation system

  • Achieved state-of-the-art performance on fashion retrieval benchmarks

• Portfolio Optimization Algorithm (Kyobo Securities)

  Jun, 2023 - Dec, 20236 months

  Developed Graph Transformer for asset price prediction and portfolio rebalancing

  • Achieved 8%+ returns in 3-month live test, outperforming benchmarks

  • Built automated rebalancing system using OHLCV data encoding

• Industrial Anomaly Detection System (N-pad)

  Sep, 2022 - Jun, 20239 months

  Created neighboring pixel-based approach for detecting manufacturing defects in industrial quality control

  • Published at CVPR Workshop 2023

  • Achieved high accuracy in detecting subtle product defects

  • Deployed in real manufacturing quality control systems

• Multi-Modal Demand Forecasting (Kyobo Book Centre)

  Mar, 2021 - Sep, 20216 months

  Built demand prediction systems for both existing and new books using deep learning

  • Created ResNet-BERT multi-modal system reducing MAE by 43.1% for new books

  • Developed Attention-LSTM system improving forecast accuracy by 38% for existing books

  • Deployed to production, replacing existing statistical forecasting system

• Automated Product Labeling Inspection (Pulmuone)

  Jul, 2021 - Oct, 20213 months

  Built automated inspection system for product labeling compliance using computer vision

  • Achieved 97% test accuracy on label verification

  • Deployed as REST API service for production use

  • Automated manual inspection process

• GNN-based Fashion Recommendation (Fashionade)

  Oct, 2020 - Feb, 20214 months

  Designed Graph Neural Network-based styling recommendation system

  • Built system to recommend items that match user's existing wardrobe

  • Integrated data from major Korean fashion platforms (Musinsa, OntheLook)

• Deep Clustering for Retail Store Segmentation (BGF Retail)

  Mar, 2020 - Jun, 20203 months

  Developed deep clustering system for convenience store segmentation based on sales patterns

  • Applied DEC and DCEC methods to cluster 12,000+ stores into 8 segments

  • Used Integrated Gradient method for explainable cluster analysis

• Attention-LSTM Demand Forecasting & Ordering System (BGF Retail)

  Jun, 2020 - Sep, 20203 months

  Created demand forecasting and ordering recommendation system using Attention-LSTM

  • Built custom loss function balancing stockout rate and inventory costs

  • Automated ordering recommendations based on target stockout rates

• Reinforcement Learning Portfolio Management (KB Securities)

  Mar, 2020 - Nov, 20208 months

  Developed portfolio rebalancing algorithm using deep reinforcement learning

  • Applied DDPG algorithm to historical stock data

  • Achieved 15% annual returns in test environment

• Medical Symptom Classification Chatbot

  Sep, 2019 - Oct, 20191 month

  Developed NLP-based medical symptom classification system for digital healthcare

  • Won Excellence Award (3rd place) at Samsung Medical Center Digital Healthcare Hackathon

  • Won Special Award (1st place in peer evaluation)

  • Achieved 91% F1 score on 10 primary symptoms and 85% on 32 dual-symptom queries

  • Trained on 50,000+ medical Q&A and 120,000+ medical news articles

• NLP-based Clinic Recommendation System

  Oct, 2019 - Dec, 20192 months

  Built symptom extraction and clinic recommendation system using NLP

  • Won Dean's Award (4th place) at SNU Medical Dream Competition

  • Won Special Award (1st place in audience evaluation)

  • Developed end-to-end system from symptom extraction to appointment booking

• Mobile Payment User Behavior Analysis (Nielsen Korea, KT)

  Oct, 2019 - Dec, 20192 months

  Analyzed cafe revisit patterns using mobile wallet transaction data

  • Analyzed 21,500 users' transaction data from 2017-2019

  • Built LightGBM and DNN multiclass classifier achieving 71% accuracy

  • Predicted revisit timing (within 7/14/21/28 days or 1+ month)

Skills

Education

Awards

Publications