Projects
* Please note that most of the projects are conducted in Korea and can be presented in Korean.
A tailored KDB-GPT model (2023-2024)
Developed a customized KDB GPT model for corporate banking, leveraging a lightweight, open-source Korean large language model (LLM). The model was implemented on-premises and trained on over 3,800 internal financial policy database tables, unstructured public data, and regulatory documents. It was designed to enhance financial decision-making by providing advanced natural language processing capabilities. Despite regulatory and security challenges posed by the bank’s internal network separation, the model was successfully deployed, significantly improving financial decision support for KDB’s corporate banking services.
Financial Data Prediction using Time-Series Analysis (2022)
Developed a financial data prediction model using time-series analysis techniques on a dataset with over 10 million rows. The model utilized Exploratory Data Analysis (EDA), data formulation, and deep learning methods, particularly Long Short-Term Memory (LSTM) networks, for time-series forecasting. Hyperparameter tuning of the LSTM model was performed to optimize its performance. Feature engineering was applied to incorporate various economic indicators, such as NASDAQ 100 and S&P 500, enhancing prediction accuracy. The project, implemented in Python, demonstrated the practical application of advanced AI and data science techniques from the AI Specialist training program.
Integrated Predictive Failure and Incident Detection System (2019)
Developed a unified monitoring system for KDB's Next Generation Banking System to address the challenge of managing 3.6M+ annual logs and ensuring 24/7 operational stability. Designed automated data collection and fault detection using 90+ queries and visualized real-time data on secure, mobile-accessible dashboards. Built LSTM-based predictive models, reducing system faults by 60.8% across 24 core systems and enhancing reliability. Integrated automated SMS alerts for rapid response, enabling proactive incident management. These efforts ensured the system's seamless deployment with zero downtime and earned the CIO Award for Excellence in 2019.
Virtualization-based In-Vehicle Infotainment System(2012)
Developed an award-winning virtualized vehicle infotainment system integrating multiple platforms (Android, MeeGo, Ubuntu) using Java, C, and Android programming. Designed a robust architecture on VirtualBox with VNC servers and clients for seamless cross-platform interaction. Integrated PulseAudio for advanced sound transmission, ensuring reliable and independent operation of core vehicle functions. This undergraduate capstone project was recognized for its innovative approach and won an Excellence Award.
Elevator Simulator based on FPGA-Based Embedded System Design(2012)
Developed an elevator simulator using VHDL to design and implement an embedded hardware system. The project was deployed on a Spartan-3 FPGA board, providing practical experience in embedded systems design and real-world applications. The simulator replicates elevator functions such as floor selection, door operations, and emergency alerts, with real-time status display via LEDs and dot matrix displays. Faced challenges in optimizing FPGA resource utilization and ensuring precise synchronization between hardware modules for accurate simulation. Overcame these obstacles through iterative testing and collaborative problem-solving within the development team.
Cloud Computing and Virtualization Research: Lessons from Silicon Valley (2010)
This study explored advancements in cloud computing and virtualization technologies, focusing on their adoption and potential applications in Korea. Investigations included visits to leading IT corporations in Silicon Valley, such as Google, IBM, and Cisco, to examine global technology trends and practices. Key objectives included identifying challenges in Korea's cloud infrastructure and proposing strategic solutions for growth. The research spanned a 14-day period (July 3–16, 2010), involving interviews with industry experts and hands-on technology demonstrations. Insights from the study aim to bridge the technological gap and enhance Korea's competitive edge in cloud services.
Sensor Network Simulation (2009)
This project simulates sensor networks to optimize energy consumption, communication efficiency, and event processing. It implements algorithms like Binary Space Partitioning (BSP) for zone allocation and the A* algorithm for efficient routing. The simulation supports various sensor distributions (Uniform, Gaussian, Clustered, Skewed) and visualizes zones, energy usage, and communication paths. Developed using C# and Windows Forms, the project features interactive elements like event queries, battery management, and hotspot analysis. It provides valuable insights into managing sensor networks under dynamic conditions.
