Research

Research Interests

Our lab is dedicated to advancing the frontier of Trustworthy, Knowledge-Grounded AI. Our core mission is to develop truthful and safe language models and AI agents that provide reliable, verifiable information to users by effectively augmenting external knowledge.

Our foundational work in Natural Language Processing spans across Knowledge-Grounded Generation, Hallucination Detection and Mitigation, and building robust Conversational AI systems.

Building on this foundation, our research agenda is unified under the goal of creating verifiably safe and aligned AI agents, from single-agent reliability to complex multi-agent collaboration. Our research is organized around three core pillars:

- Foundational Language Intelligence: Context Engineering and Memory

We focus on the core cognitive capabilities of language agents - how they acquire, manage, and reason with information to build a reliable understanding of the world. We are pioneering advanced context engineering techniques to ensure that agent actions are grounded in accurate and relevant knowledge.

Main Topics: Advanced Agentic RAG, Agentic Memory

- Verifiable Safety and Policy Alignment in Language Agents

We address the critical challenge of ensuring that language agents operate safely and in accordance with human values and explicit rules. Our goal is to make agent behavior not just reliable but provably safe by leveraging our deep expertise in NLP safety.

Main Topics: Policy Adherence & Alignment, Safe Agent Design & Evaluation

- Collaborative Intelligence in Multi-Agent Systems

We extend our language-centric research from single agents to the complex, interactive dynamics of multi-agent systems. We aim to lead the development of frameworks that ensure these collaborative systems are effective, safe, and aligned.

Main Topics: Multi-Agent Orchestration, Safety for Multi-Agent Systems, Role-Playing Agents

Collaborators

Our group encourages collaboration with researchers from other institutions and research groups to widen the horizon of our research capability. We are undertaking a project or will collaborate with the following institutes.