Skip to content

Latest commit

 

History

7 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 

Repository files navigation

👋 Chaewon Lee입니다!

NLP · LLM · Multimodal Learning 중심으로 문제를 푸는 AI 연구자


Profile

  • B.S. in Industrial & Data Engineering, Hongik University (Mar. 2022 – Feb. 2027)
  • Undergraduate Research Assistant, Artificial Intelligence Application Lab, Hongik University
  • Undergraduate Research Assistant, Intelligent Robot Optomechatronics Lab, DGIST
  • Data Evaluation Intern, Kanana Quality Assessment Team, Kakao Corp.
  • Research interests: Computer Vision, Multimodal AI, Medical AI, Human-Centered AI

Research Interests

  • Computer Vision (3D Reconstruction, Detection, Segmentation)
  • Multimodal Learning (Image · Signal · Text)
  • Medical AI & Healthcare Applications
  • Human-Centered AI & Model Evaluation

🧪 Research Experience

3D CT Reconstruction from 2D X-ray Images

Artificial Intelligence Application Laboratory, Hongik University
Sep. 2024 – Dec. 2024

  • Generated DRR-based synthetic X-ray images using LIDC-IDRI
  • Constructed 2-channel inputs by combining frontal and lateral views
  • Implemented 3D U-Net-based Generator and Patch-based Discriminator
  • Improved volumetric CT reconstruction quality by combining L1 Loss + GAN Loss

AI-Based Puncture Prediction from Optical Signal Data

Intelligent Robot Optomechatronics Laboratory, DGIST
Jul. 2025 – Aug. 2025

  • Modeled puncture prediction from 5-channel optical time-series signals
  • Explored LSTM, zero-shot, and few-shot approaches
  • Achieved best performance with ResNet18 (AUC 0.955) through extensive tuning
  • Addressed class imbalance and early prediction issues via loss redesign and sampling strategies

AI-Based Needle Safety Classification & Red Line Detection

Intelligent Robot Optomechatronics Laboratory, DGIST
Jul. 2025 – Aug. 2025

  • Detected red reference lines in X-ray images using HSV-based color filtering
  • Identified needle positions with YOLOv8s
  • Built an MLP classifier using geometric features with ensemble learning
  • Achieved F1-score 0.965, Accuracy 0.951, and 98.6% precision for high-risk detection
  • Extended binary classification to 4-class risk level prediction

Industry Experience

Data Evaluation Intern

Kanana Quality Assessment Team, Kakao Corp.
Sep. 2025 – Dec. 2025

  • Designed HFL-based evaluation test cases for the LLM Kanana
  • Improved test case quality metrics by approximately 8%
  • Categorized test cases by factuality, multi-step reasoning, and error handling
  • Conducted benchmarking against Gemini and GPT
  • Reduced inter-evaluator variance by formalizing evaluation guidelines and reference answers

💼 Project Experience

🧾 LLM-Based Medical Receipt Document QA & RAG Agent

Jan. 2026 – Feb. 2026

  • Built an LLM-powered document understanding and QA pipeline for medical receipt images
  • Implemented hybrid RAG (BM25 + vector retrieval) for evidence-grounded responses
  • Orchestrated an end-to-end LangChain-based agent with an interactive Streamlit interface

AI-Based Bankruptcy Prediction & Risk Profiling System

Sep. 2024 – Dec. 2024
🔗 Streamlit Demo: https://hongik1.streamlit.app/

  • Developed bankruptcy prediction models (DNN, Decision Tree, KNN)
  • Achieved AUC 0.99, Recall 0.97 with DNN
  • Applied KMeans + PCA for high-risk customer segmentation
  • Identified key risk drivers using Random Forest
  • Deployed real-time inference via Streamlit

Academic Paper (In Preparation)

Optimizing Drone Station Locations and Drone Allocation Using Forecasted Demand for Emergency Blood Pack Delivery
Lee, C., Jung, H., Jang, H., Yu, J., Lee, C., & Kim, H. L.

  • Recipient of Council President’s Award
  • CO-Data Station Essay Competition, National Research Foundation of Korea

Extracurricular Activities

Prometheus – Artificial Intelligence Study Group

Mar. 2025 – Aug. 2025

  • Developing YOLOv8-based multi-food object detection models
  • Experimenting with YOLOv5 / YOLOv8 for performance comparison
  • Implementing GNN-based nutrient matching algorithms

Data-Driven Optimization for Emergency Medical Drone Delivery

Mar. 2024 – Aug. 2024

  • Designed an end-to-end framework integrating demand forecasting and facility location optimization
  • Modeled emergency blood demand using public medical and traffic datasets
  • Applied K-means clustering to hospital geospatial data with Q95 demand indicators
  • Solved a location-allocation optimization problem under operational constraints
  • Conducted simulation-based validation for scalability analysis

Tech Stack

Languages & Tools
Python, R, Git, Jupyter, VS Code, Streamlit, Notion

Frameworks & Libraries
PyTorch, TensorFlow, scikit-learn, Pandas, Matplotlib

AI / ML
Computer Vision, Deep Learning, Time-Series Modeling, Multimodal Learning, Optimization


Certifications & Awards

  • TOEIC Speaking IH (2025)
  • SQLD (2024)
  • ADsP (2023)
  • Merit Scholarship, Hongik University (2024–2025)
  • Council President’s Award, CO-Data Station Competition (Top 7 Nationally)

💻 Tech Stack

Python R PyTorch TensorFlow scikit-learn Pandas Matplotlib Jupyter Git Streamlit VS Code Notion


📫 Contact


About

No description, website, or topics provided.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors