Rigorous Task Formulation
I define concrete research problems from underexplored and messy real-world settings by clarifying assumptions, designing tasks, setting baselines, and building evaluation criteria.
KAIST AI Ph.D. candidate - graph mining, retrieval, agentic AI
Structure-Aware AI Researcher
I study how complex relationships, dependencies, and reusable knowledge in real-world data can be represented, modeled, and used in AI systems. My research builds on graph and hypergraph methods, with recent extensions toward knowledge-structured memory and retrieval-augmented AutoML agents.
Research strengths
Turning Hidden Structure into Concrete ML Research
I find non-obvious structure in complex problems, formalize it as graph-based data or ML tasks, and carry it through modeling, experiments, and publication.
16
publications across KDD, WWW, ICDM, IJCAI workshops, AAAI workshops, ECML/PKDD workshops, and TKDD
6
first-author or co-first-author works
1
Best Paper Award at IEEE ICDM 2025
1
industry collaboration published at KDD Industry Track
Structure-Aware Modeling
I read complex data as entities, relations, groups, and higher-order interactions, then turn them into graph, hypergraph, or knowledge-graph representations.
Persistent System Building
I execute research end-to-end by building models and multi-agent systems that connect structured knowledge to planning, coding, debugging, and feedback.
Current research
Graph memory for multi-agent AI.
My current direction explores how graph and hypergraph structures can serve as memory for multi-agent AutoML systems. The focus is on preserving useful workflow structure while allowing agents to retrieve, adapt, and recombine prior knowledge for new tasks.
Research direction
Hypergraph-structured memory for agentic AutoML
I am applying my hypergraph research background to agent memory: representing reusable technique blocks, workflow dependencies, and execution feedback as structured knowledge that multi-agent systems can use during planning and experimentation.
Experience basis
Why this direction fits my record
Hypergraph foundation: problem formulation, analysis, sampling, generation, and learning for group-level interaction data.
Graph-structured retrieval: experience with large-scale search and graph-based structure, including satellite video retrieval for forecasting.
Multi-agent systems: direct NEC experience building planner-retriever-coder-debugger systems for ML automation.
Publications
Full research record.
My publications cover higher-order network modeling, graph learning, anomaly detection, temporal networks, large-scale retrieval, and AI system applications.
2026
TacitFlow: Learning Workflow Representations for Tacit-Knowledge-Grounded Machine Learning Engineering Agents
KDD 2026 AI Data Scientist Workshop
2025
Attributed Hypergraph Generation with Realistic Interplay Between Structure and Attributes
IEEE ICDM 2025 Best Paper Award
2025
2025
Kronecker Generative Models for Power-Law Patterns in Real-World Hypergraphs
The Web Conference (WWW) 2025
2025
Identifying Group Anchors in Real-World Group Interactions Under Label Scarcity
IEEE ICDM 2025
2024
Representative and Back-In-Time Sampling from Real-world Hypergraphs
ACM Transactions on Knowledge Discovery from Data (TKDD)
2024
GraphEHR: Heterogeneous Graph Neural Network for Electronic Health Records
IJCAI 2024 Workshop
2024
2024
2023
2023
2022
2022
2022
2021
How Do Hyperedges Overlap in Real-World Hypergraphs? - Patterns, Measures, and Generators
The Web Conference (WWW) 2021
2020
Pretraining Neural Architecture Search Controllers with Locality-based Self-Supervised Learning
NeurIPS 2020 Workshop
Experience
Research roles and collaborations.
Each experience contributed a different layer to my current work: knowledge structuring, multi-agent systems, graph modeling, and scalable retrieval systems.
Samsung
Industry-linked research project
Scientific evidence reasoning for R&D support
Identified scientific QA challenges where answers depend on conditions, context, and evidence across papers. Built paper-grounded evaluation data and applied multi-agent verification to improve grounding, answer support, and reliability.
NEC Laboratories America
Research Intern - Data Science & System Security
Time-series retrieval and multi-agent AutoML
Used time-series foundation models for multi-view similar-case retrieval and evidence-format analysis. Built a Planner-Retriever-Coder-Debugger AutoML agent with structured memory for feedback-driven ML pipeline improvement.
Carnegie Mellon University
Visiting Researcher
Compact generative modeling for real-world hypergraphs
Collaborated with Prof. Christos Faloutsos on principled hypergraph generative modeling, resulting in a WWW 2025 publication on reproducing real-world higher-order network patterns.
Korea Meteorological Administration
Joint project
Similar past-case retrieval for weather forecasting
Built self-supervised satellite video retrieval to help forecasters find similar past weather cases from large historical archives. The deployed forecasting support system led to a KDD 2025 Industry Track publication as co-first author.
Background
Education, skills, and service.
Education
KAIST, M.S. and Ph.D. in Artificial Intelligence, advised by Prof. Kijung Shin. Expected February 2027.
Education
Sogang University, B.S. in Computer Science and Engineering. GPA 4.14/4.3, Summa Cum Laude, ranked 1st of 72.
Skills
Graph and hypergraph mining, knowledge graph construction, LLM agent systems, retrieval-augmented AI, PyTorch, PyTorch Geometric, DGL, Python, C/C++, Java.
Service
Co-organizer, PAKDD 2025 Workshop on Graph Learning with Foundation Models. Reviewer for ACM KDD, WWW, ACM CIKM, and PAKDD.
Awards
IEEE ICDM 2025 Best Paper Award, WWW 2021 Student Scholarship Award, Grand Prize in Open-Source Software Contribution, National Science and Technology Scholarship.