ADMIT: Support-Gated Memory-Write Admission for Document QA Agents
Support-gated admission control that decides what a document QA agent is allowed to commit to memory.
Publications
My publications investigate how AI capabilities can translate into useful and reliable systems in practice. Across industrial RAG, multimodal reasoning, evidence grounding, and agent memory, I often use explicit structure to preserve information that matters for downstream use. * denotes first or co-first author. See also Google Scholar.
Conference Papers 6
PILAR: A Page-Grounded Unified Evidence Representation via an Entity-Linked Assertion Graph for Open-Domain QA Agents over Multimodal Document Corpora
Entity-linked assertion graphs unify text, tables, and figures into a single page-grounded evidence representation for open-domain multimodal QA agents.
HiKEY: Hierarchical Multimodal Retrieval for Open-Domain Document Question Answering
Hierarchical retrieval for multimodal document QA with structured evidence assembly.
M3DocDep: Multi-modal, Multi-page, Multi-document Dependency Chunking with Large Vision-Language Models
LVLM-based dependency chunking that reconstructs cross-page structure for long-document retrieval and QA.
MultiDocFusion: Hierarchical and Multimodal Chunking Pipeline for Enhanced RAG on Long Industrial Documents
A hierarchical multimodal chunking pipeline that preserves layout and improves evidence composition in industrial RAG.
Intelligent Predictive Maintenance RAG Framework for Power Plants: Enhancing QA with StyleDFS and Domain Specific Instruction Tuning
Domain-specific RAG framework for scientific and industrial QA; led to two technology transfers.
Journals 2
Domestic Conferences & Theses 9
-
Award: Outstanding Paper Award
-
Comparative Analysis of Korean Quality in Large-Scale Language Models Based on Zero-Shot Learning
-
A Neural-Symbolic Model for Overcoming Deep Learning Limitations in Korean Dependency Parsing
-
EDT5: Proposed Embedding Model of T5 Encoder-Decoder Structure
Award: Outstanding Paper Award
-
Award: Student Research Award
-
Korean Machine Reading Comprehension Using Continual Learning
Manuscripts Under Review 4
4 first-author manuscripts under peer review. Author lists are withheld under double-blind review.
First author
-
Evaluating the Practical Utility of Agentic Memory
-
Citation Grounding for Video Question Answering
-
Evidence Auditing and Support-Sensitive Evaluation for Multimodal QA
-
Error Propagation Diagnostics for PDF-to-RAG Pipelines Across Representation Families
Patents 5
-
Hierarchical Multimodal Retrieval for Open-Domain Document Question Answering
-
Multi-modal, Multi-page, Multi-document Dependency Chunking with Large Vision-Language Models
-
Device and Method for Document Chunking, and Question-Answering Device and Method Using the Same
-
Rule Filtering Techniques and Methods for Knowledge Inference Systems Based on Deep Learning
-
AI-Based Adaptive Problem Recommendation System and Method for Learner-Level Assessment
Current research directions include actionable ontologies for multimodal agent memory, video-structured memory for long-horizon reasoning, and planning over structured world representations.