Nguyen (William) Nguyen

Currently, I am a Senior Applied Scientist at Aitomatic. I graduated with a master's degree from the University of Rochester, where I was advised by Professor Chenliang Xu. During my time at the University of Rochester, I worked on instructional video understanding and vision-language problems.

I had nearly 3 wonderful years at VinAI Research where I worked on the scene text spotting problem under the supervision of Professor Nguyen Minh Hoai. I completed my bachelor's degree at the University of Engineering and Technology - VNU, where I had the opportunity to work with Professor Hoang Van Xiem.

Email  /  CV  /  LinkedIn  /  Bio  / 
Google Scholar  /  Twitter  /  Github
✨ Chat with my AI Twin ✨

I'm interested in natural language processing, computer vision, and machine learning (especially deep learning). Currently, I am working on developing domain-specific foundation models and multi-agent systems for industry applications. My recent work focuses on building intelligent AI agents that can solve complex problems through adaptive reasoning and exploration, as well as creating specialized large language models for domains like semiconductors and maritime. Previously, I worked extensively on optical character recognition and vision-language problems, focusing on scene text recognition and understanding visual content using natural language.

A modular, general-purpose multi-agent framework for iterative problem solving through exploration and plan evolution. ProSEA features a hierarchical architecture with a Manager Agent orchestrating domain-specialized Expert Agents, enabling adaptive replanning based on structured feedback from failures and newly discovered constraints. Demonstrates state-of-the-art performance on FinanceBench benchmark.

The first opensource domain-specific LLM for the maritime industry. Our model outperforms serveral commercial products, including GPT-4o-mini GPT-4o, Claude-3.5-Sonnet, and opensource models such as Llama3.1 8B, Llama3.1 70B, and Llama3.3 70B.

The first opensource domain-specific LLM for the semiconductor industry. Our model outperforms serveral commercial products, including Claude-3.5-Sonnet, Haiku, Opus, Command-R, and opensource models such as Llama3 70B.

An unified agentic framework to incorporate expert knowledge using neurosymbolic to create domain specific agents that can perform with high consistency and accuracy, significantly outperform ChatGPT assistant and ReAct agent.

An unified multimodal LLM designed for comprehensively solving all tasks related to egocentric video understanding.

A new task of understanding the state of objects and the progression of object state changes. Besides, we trained a Multimodal-LLM that significantly surpasses previous state-of-the-art models. Our model achieved 90% quality of GPT4-V on both GPT4 and human evaluations.

A simple but effective approach to incorporate language knowledge from large text corpus for improving both text detection and recognition.

A comprehensive system designed to guide humans to work more efficiently and accurately. Our system leverages the power of LLMs to interpret and process information from visual, auditory, and contextual dimensions.

A novel approach to incorporate dictionary on both training and testing phases. Additionally, we also introduced a novel Vietnamese scene text dataset (VinText), the largest scene text dataset for Vietnamese.

• Delivering Domain-Expert Agents and Models Using Synthetic Knowledge
  Christopher Nguyen, Manh-Nguyen Nguyen, Hong An Phan, Zooey Nhu-Quynh Nguyen, The-Vinh Luong, Elise NhuY Nguyen, Thomas Rasmussen, Anh Hai Ha, Phi-Hung Vo, Xuan-Sang Dinh, Huy-Thuan Bui, Anh-Quoc Dang, Timothy Michael Gerard Rozario
  US Patent App. 63/726,322, 2024.
• Delivering Domain-Expert Agents for Improving Problem-Solving
  Christopher Nguyen, Manh-Nguyen Nguyen, Hong An Phan, Zooey Nhu-Quynh Nguyen, The-Vinh Luong, Elise Nhu-Y Nguyen, Thomas Rasmussen, Anh Hai Ha, Phi-Hung Vo, Xuan-Sang Dinh, Huy-Thuan Bui, Anh-Quoc Dang
  US Patent App. 63/721,419, 2024.
• Domain-Aware Neurosymbolic Agents For Improving Problem-Solving Accuracy And Consistency
  Christopher Nguyen, The Vinh Luong, Xuan Sang Dinh, Zooey Nhu-Quynh Nguyen, Shruti Raghavan, Manh-Nguyen Nguyen, Quynh Thi-Tham Le, Phi Hung Vo, Tan Loc Nguyen, Anh Hai Ha, Phuong Thao Nguyen
  US Patent App. 63/696,337, 2024.

State-aware Object Understanding
[Master's Thesis]

Reviewer
WACV 2022, CVPR 2023, CVPR 2024, ACMMM 2024, AAAI 2025, NAACL 2025, CVPR 2025, ACL 2025

Competition jury member (2021)
Ho Chi Minh city AI Challenge 2021: Vietnamese Scene Text Recognition

Teaching assistant (2018 - 2019)
University of Engineering and Technology - Vietnam National University
Teaching assistant in several computer vision and machine learning courses for Samsung Display Vietnam's staff