MSc student & AI researcher · Abu Dhabi, UAE
I am an MSc student at Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), where I am supervised by Associate Professor Salman Khan. Before joining MBZUAI, I worked at the AISIA Research Lab with Associate Professor Binh T. Nguyen and collaborated with Assistant Professor Lizi Liao at Singapore Management University. My research spans multimodal learning, natural language processing, video retrieval, trustworthy emotional-support systems, Vietnamese language modeling, medical image analysis, and scientific AI. I am particularly interested in practical AI systems that coordinate external tools, reason across heterogeneous data, and keep their conclusions traceable to evidence.
Research outlook
I am currently interested in developing reliable, long-horizon AI agents that can reason over structured trajectories, use external tools, detect and recover from failures, and keep their decisions grounded in verifiable evidence. In particular, I am exploring process-level representations, post-training, and reinforcement learning methods for improving agent planning, replanning, memory, and recovery across scientific workflows. For 2027, I am open to research internships and collaborations in agentic AI, reasoning and post-training, reinforcement learning, multimodal and scientific agents, trustworthy AI, and AI for science.
Updates
Joined MBZUAI as an MSc student.
Released TerraBench, a benchmark for grounded Earth-science reasoning with executable agents.
Published MULTIMOOD at AAAI on trustworthy multimodal emotional-support systems.
Joined MBZUAI as a research engineer.
Published ViWordFormer at AAAI, rethinking word segmentation for Vietnamese NLP.
Research
Representative figure from this publicationMULTIMOOD combines video, audio, text, and dialogue history to recognize emotional-support needs and generate responses aligned with professional therapeutic criteria.
Representative figure from this publicationA benchmark and executable agent framework for grounded reasoning across Earth observation, gridded environmental data, GIS, simulation, and scientific documents.
Representative figure from this publicationA faster, more accessible video-retrieval system with optimized keyframe extraction, multilingual OCR and ASR, lightweight visual question answering, and interactive reranking.
Representative figure from this publicationA linguistic and empirical re-examination of Vietnamese word segmentation, with ViWordFormer modeling phrasal lexemes directly from Vietnamese text.
Representative figure from this publicationAn evaluation of six deep-learning models for grading visual clarity in shoulder arthroscopy, supporting more consistent assessment of bleeding-related image quality.
