CV

Joowan Chae

Sungkyunkwan University, Suwon, South Korea
B.S. in Software (Double Major) and Bio-Mechatronic Engineering
Mar. 2021 - Present

Data Intelligence and Analytics Lab (DIAL), SKKU
Undergraduate Research Program Trainee
Advisor: Prof. Jongwuk Lee
Mar. 2026 - Present

Participating in the KDD Cup 2026 international competition as a key milestone of the URP, focusing on autonomous data agents for complex data analysis.
Designed a context-aware schema manifest system that analyzes CSV, JSON, SQLite, and Markdown files to extract question-relevant tables, columns, values, and domain rules.
Implemented a lightweight Markdown-based knowledge retrieval module to connect code definitions, thresholds, and domain-specific rules to structured data.
Extended the ReAct agent loop with difficulty-aware planning, self-correction, loop detection, and candidate answer verification.

Safe-Ticket
Capstone Design Project
Mar. 2026 - Present

Developing a real-time fraud detection web service and Chrome extension for secondary ticket trading markets.
Implemented Retrieval-Augmented Generation (RAG) and semantic search algorithms to identify complex fraud patterns and streamline evidence packaging.

News Summarization Model Fine-tuning
Independent Project
Jan. 2026 - Feb. 2026

Fine-tuned a Qwen 2.5 0.5B language model using Chain of Density prompting and Odds Ratio Preference Optimization (ORPO).
Evaluated downstream summarization performance using BERTScore and HHEM to support concise and factually accurate generation.

Programming Languages: Python, C, C++
AI and Machine Learning: LLM alignment, LLM reasoning, agentic workflows, RAG, semantic search
Alignment Methods: DPO, KTO, SimPO, ORPO, GRPO
Frameworks and Tools: LangChain, LangGraph, Git

Independent Literature Reviews on Foundation Models and Alignment
Sep. 2025 - Present

LLM Alignment and Preference Optimization: RLHF, PPO, DPO, KTO, SimPO, ORPO, reward modeling, reference-free optimization, stability, and data efficiency.
Advanced Reasoning and Agentic Problem Solving: ReAct, rStar-Math, process reward models, ReFT, AOP, ACT-SQL, mathematical reasoning, Text-to-SQL, and tool-augmented orchestration.
Representation Learning and Embedding Models: LLM2Vec, Qwen3 Embedding, M3-Embedding, contrastive learning, dense retrieval, reranking, multilingual embeddings, and RAG.
Foundation Model Architecture and Pre-training: Transformer-XL, XLNet, ELECTRA, DeBERTa, RoFormer, MASS, LoRA, long-context modeling, and parameter-efficient fine-tuning.
Mathematical Foundations of ML: singular value decomposition, matrix factorization, eigenvalues, eigenvectors, and gradient descent variants.