Projects
Conversational Agents with Hyper Long-Term Memory 초장기기억 기반 대화형 에이전트
- Led research on hyper long-term memory for conversational agents, focusing on persistent preference tracking and multi-session personalization.
- Designed a preference-aware memory framework that links long-horizon user behavior to structured tool-calling decisions.
- Built end-to-end pipelines for long-context data construction, memory-grounded modeling, and evaluation under realistic multi-session settings.
- Developing robust methods for preference abstraction and retrieval to improve consistency, controllability, and practical deployment of proactive dialogue agents.
Multi-Intent Modeling for Advanced Speech Recognition 음성인식 고도화를 위한 복합의도 모델 연구
- Led foundational research on transition-aware multi-intent conversational AI, resulting in an EMNLP 2025 main conference publication.
- Designed and developed a preference-aligned dialogue framework integrating dataset construction, hybrid LLM-based validation, and multi-intent reasoning.
- Owned the end-to-end research and development pipeline, from modeling and data infrastructure to deployment of an LLM-based input analysis module for in-vehicle voice assistants.
- Directed project execution in an industry-academia collaboration, coordinating research contributors, managing milestones, and delivering deployable system outcomes.
Natural Language-Based Compound Intent Algorithms 자연어 기반 복합 의도 알고리즘 연구
- Led research on multi-intent understanding for speech-based dialogue systems, addressing limitations of single-intent voice interaction models.
- Initiated and directed the construction of a large-scale multi-intent conversational dataset designed to model complex and transition-aware user interactions.
- Designed and led the development of a speech utterance segmentation-based multi-intent detection algorithm enabling accurate identification of multiple intents within a single user query.
- Resulted in a long paper publication at LREC-COLING 2024 and domestic and international patent applications derived from the proposed dataset construction and modeling methodology.
- Oversaw end-to-end research execution, organizing datasets, algorithms, and experimental frameworks into reproducible research artifacts.