I am a undergraduate student at Yonsei University majoring in Computer Science. Currently, I'm tackling challenges in reinforcement learning and robotics, advised by Prof. Youngwoon Lee.
Prior to this, I worked as a research intern at LangAGI lab, conducting research on LLM-based web agents.
Publications
Web Agents with World Models: Learning and Leveraging Environment Dynamics in Web Navigation
Hyeongjoo Chae, Namyoung Kim, Kai Tzu-iunn Ong, Minju Kwak, Gwanwoo Song, Jihoon Kim, Seonghwan Kim, Dongha Lee, Jinyoung Yeo
Preprint
We propose a World-model-augmented (WMA) web agent for improved decision-making in long-horizon tasks. Our study reveals that current LLMs lack world models, leading to suboptimal performance in web navigation. To address this, we introduce a transition-focused observation abstraction method, which uses natural language descriptions to highlight key state changes between time steps. Our emperical results show that our world models enhance agents' policy selection without additional training, offering superior cost- and time-efficiency compared to recent tree-search-based agents.
Projects
AI Agent for Finance Risk Detection
YBIGTA x UPSTAGE AI Agent Hakathon 2025
We developed a RAG-based agent that helps users identify potential risks in financial products. To support this, we collected financial dispute cases and legal documents to build a structured database. When a user inputs a financial product, the agent highlights potentially risky sentences in a PDF viewer. Additionally, users can ask follow-up questions, such as simulations of possible future outcomes or inquiries about related financial information.
Visualizing Global Health Risks: Smoking, Drinking, and Cancer Rates
2024-1 Data Visualization Course Project
We collected and visualized time-series data on organ-specific cancer incidence, smoking, and alcohol consumption across countries. The interface includes an interactive human body diagram—when a user clicks on an organ, related scatter plots and line charts appear, showing cancer incidence alongside relevant lifestyle factors. For visualizing country-level trends, we used a grid cartogram to present the data in an interactive and intuitive way.
University Chatbot and Course Search Engine
AWS-YONSEI 2023 Winter Program
We developed a chatbot using data from a university community platform and a course search engine based on Yonsei University's course catalog. The chatbot leverages RAG with vector similarity search to provide appropriate responses to users. The course search engine achieves fast inference by building an inverted index for efficient retrieval.
Services
Reviewer
- System-2 Reasoning Workshop, NeurIPS 2024