Responsible AI Researcher, Frontier AI Lab, KT Corporation · Adjunct Professor in Business Analytics, UNIST
AI safety, governance, and value alignment: red-teaming and defence of retrieval-augmented generation systems, risk taxonomies for high-risk and Physical AI, and evaluation of large language model judges.
Full profile, publications, and CV: deep1003.github.io
RAI-Risk-Taxonomy-2.0 · HTML Evidence-linked responsible AI risk taxonomy spanning general-purpose, agentic, and Physical AI. 1,726 global L4 risks across 50 L3 nodes, with the RAI-HOLD review queue and reproducible reliability validation. Site · Technical report · Semantic space
Physical-AI-Risk-Taxonomy · HTML Interactive L4 risk-card taxonomy for Physical AI: embodied AI, humanoids, robots, drones, autonomous vehicles, cyber-physical systems. 182 risk cards with bilingual definitions, evidence justifications, curation data, and audit logs. Site · Expert survey
Physical-AI-Risk-Survey-Responses Anonymous expert annotations collected for the Physical AI Risk Taxonomy.
moral-orientation-calibration · Python Minimal reproducibility release for Moral Orientation and Calibration: Coupled in Human Annotators, Separable in Judge LLMs, ICML 2026 Pluralistic Alignment Workshop. Paper
silent-failure-moral-alignment Reproducibility release for Silent failure of moral alignment in large language-model evaluators, under review at Scientific Reports.
study-fbi-survey · HTML Public GitHub Pages frontend for the Study FBI human questionnaire on context-dependent aesthetic judgment. Anonymous server-side response collection. Questionnaire
Related paper, code hosted elsewhere: KidnapRAG — black-box attack hijacking reasoning in agentic RAG systems. Code
patstat-online-ai-patent-collector · Jupyter Notebook Reproducible notebook workflow for year-specific AI patent source-data collection from PATSTAT Online under query-cost, query-length, and download-size constraints. Automates six relational query groups into a one-row-per-patent master file. Output dataset: PATSTAT AI Complete Master, 1950–2026 — 2,330,553 applications, 175 variables.
dci-patent-atlas · Python Data Center Interconnect patent atlas: science, technology, and supply-chain mapping.
AI_Topic_Space.github.io · HTML Interactive semantic map of AI topics across policy reports, academic papers, and patents. Four-level topic hierarchy with Korea activation and relative topic-gap analysis. Site · Dataset
tim510-preclass-survey · HTML Pre-class survey for TIM510 Business Analytics, Graduate School of Technology and Innovation Management, UNIST.
deep1003.github.io · HTML Source for my academic homepage.
deep1003 · Jupyter Notebook This profile repository, plus teaching and research notebooks on patent analytics, Sentence-BERT, and PatentSBERTa.
[email protected] · Homepage · Google Scholar · ORCID · LinkedIn