About Me

I am an AI researcher working on reliable multimodal agents: systems that can perceive visual environments, reason over time, and take grounded actions. My work spans vision-language models, LLM reasoning and planning, and multimodal evaluation, with a focus on visual grounding, temporal video understanding, and reducing hallucinations.

At the University of Maryland and UMBC, I have worked on VLM hallucination mitigation, trace-grounded video reasoning, executable planning from natural language, code reasoning, and persona-aware conversational agents. I also have applied engineering experience building multimodal RAG systems and agentic workflow integrations.

My long-term goal is to develop agents that combine vision, language, memory, and planning to complete reliable long-horizon tasks — first in digital environments and eventually in embodied settings.

Research Interests

Multimodal perception and grounding: Vision-language models, video understanding, temporal reasoning, visual grounding, and hallucination detection and mitigation.

Reasoning, planning, and verification: LLM agents, state tracking, executable planning, code reasoning, and methods that let agents check and correct their own outputs.

Memory, world models, and action: Long-horizon multimodal memory, world models, visual-context compression, vision-language-action models, and agents for GUIs and embodied environments.

Broadly, I am interested in AI systems that know when to look, what to remember, how to reason, when to verify, and how to act.

News

• [Academic] August 2026: Paper "Frames2LoRA: Parametric Video Internalization for Vision-Language Models" accepted at EMNLP Findings 2026!
• [Academic] August 2026: Paper "PlanForge: Synthesizing Planners from Natural Language with an LLM-Driven Architect–Builder–Runner Pipeline" accepted at the COLM 2026 Workshop on Efficient Reasoning!
• [Academic] March 2026: Research paper "Towards Mitigating Hallucinations in Large Vision-Language Models by Refining Textual Embeddings" accepted at ACL Findings 2026!
• [Academic] March 2026: Paper "The Low-Frequency Trap: Why Scaling Doesn't Solve Simple Temporal Counting" accepted at the ICLR 2026 Workshop!
• [Academic] February 2026: Co-authored paper "Analyzing Chain of Thought (CoT) Approaches in Control Flow Code Deobfuscation Tasks" accepted at the WoRMA 2026 Workshop!
• [Academic] May 2025: First-author paper "(CPER) From Guessing to Asking" accepted at NAACL SRW 2025!