Tin (Kevin) Nguyen

๐ŸŽ“ I am a PhD student in Computer Science at Auburn University, working with Prof. Anh Totti Nguyen and Prof. Chirag Agarwal at UVA.
My research focuses on interpretable-by-design, agentic AI, and HCI.

I'm looking for postdoc, internship, and full-time opportunities around LLMs, VLMs, Agents ๐Ÿ‘€ Please email me if you think I'd be a good fit: ngthanhtinqn@gmail.com or ttn0011@auburn.edu ๐Ÿค—

Education

๐ŸŽ“ Auburn University
Ph.D. in Computer Science and Engineering
AL, U.S. · Aug. 2022 – present
Advisor: Anh (Totti) Nguyen
Teaching Assistant: Formal Languages, Machine Learning, Security
๐ŸŽ“ Sejong University
M.Sc. in Computer Science and Engineering
Seoul, S. Korea · 2020 – 2022
Advisor: Kim Yong-Guk
๐ŸŽ“ University of Science
B.A. in Computer Science and Engineering
Ho Chi Minh, Vietnam · 2015 – 2019
Advisor: Ngoc Quoc Ly

Experience

๐Ÿ’ผ Aikyam Lab, University of Virginia
Graduate Research Assistant
VA, U.S. · May – Aug. 2026
Advisor: Chirag Agarwal · Project: Multimodal Agentic Models + Trustworthy AI
๐Ÿ’ผ AIOZ Co., Ltd.
AI Engineer
Ho Chi Minh, Vietnam · 2019 – 2020
Built a license plate recognition system and a ball tracking system
๐Ÿ’ผ Gumi Co., Ltd.
AI Engineer Intern
Ho Chi Minh, Vietnam · 2018 – 2019
Built Japanese character recognition for bills and documents

Honors & Awards

๐Ÿ† Tinker Research Grant
Aug. 2026
Tinker · Awarded $5,000 for compute/API credits
๐Ÿ† GSC Travel Fellowship
Dec. 2025
Auburn University Graduate School Council · Awarded $400 to support conference travel
๐Ÿ† Tinker Promotion
Nov. 2025
Tinker · Awarded $150 in credits
๐Ÿ† 1st Prize, AICovidVN Challenge
2021
Ho Chi Minh, Vietnam · Introduced Fruit-CoV, a framework detecting SARS-CoV-2 through cough sounds (~$4,300 prize)

Publications my favorites () | others ()

PageGuide paper figure
Agent Human-Computer Interaction
Tin Nguyen, Thang T. Truong, Runtao Zhou, Trung Bui, Chirag Agarwal, Anh Totti Nguyen
Arxiv, 2026
PageGuide is a browser extension that acts as an in-page navigation agent: given a user's question, it locates the supporting evidence on the current webpage and highlights exactly where the answer comes from, instead of leaving the user to skim the whole page. We found that this agent improves the users verification up to 15%, post-survey questions show that users prefer evidence to verify agent's answer rather than plain text answers from non-grounding web agents (Claude Ext, Gemini Ext, etc).
HoT paper figure
Prompt Engineering Human-Computer Interaction
Tin Nguyen*, Logan Bolton*, Mohammad Reza Taesiri, Trung Bui, and Anh Totti Nguyen
TMLR, 2026
Also accepted at NeurIPS 2025 Workshop Multimodal Algorithmic Reasoning (MAR) โ€” Oral
HoT reformats an LLM's chain of thought so every claim is directly highlighted and linked back to the exact supporting facts in the input, using matched color-coded spans between the reasoning and the source text. This lets a reader quickly check whether the model's reasoning is actually grounded in the input rather than hallucinated, and the paper shows it improves both faithfulness and human trust compared to standard chain-of-thought prompting.
PEEB paper figure
Interpretable-by-Design Human-Computer Interaction
Thang Pham*, Peijie Chen*, Tin Nguyen*, Seunghyun Yoon, Trung Bui, Anh Nguyen
NAACL, 2024 Findings
PEEB is a part-based, interpretable-by-design image classifier that predicts an object's class by first describing its visible parts in natural language (e.g., wing color, beak shape) and matching those descriptions against a text-conditioned part detector. Because the classification bottleneck is human-readable text, users can inspect exactly which part attributes drove a prediction and edit the class descriptions to correct or adapt the model without retraining.
VizDoom navigation figure
Vision-Language-Action
Thanh Tin Nguyen*, Anh H. Vo*, Soo-Mi Choi, Yong-Guk Kim
Knowledge-based Systems (KBS), Jul 4, 2023
This work studies vision-and-language navigation in 3D environments built on VizDoom, where an agent must follow a natural-language instruction to reach a target location. The proposed coarse-to-fine fusion combines instruction and visual features at multiple stages, first aligning coarse landmarks mentioned in the instruction and then refining fine-grained action decisions, which improves navigation success over single-stage fusion baselines.
Maijunxian Wang, Ruisi Wang, Juyi Lin, Ran Ji, Thaddรคus Wiedemer, Qingying Gao, Dezhi Luo, Yaoyao Qian, Lianyu Huang, Zelong Hong, Jiahui Ge, Qianli Ma, Hang He, Yifan Zhou, Lingzi Guo, Lantao Mei, Jiachen Li, Hanwen Xing, Tianqi Zhao, Fengyuan Yu, Weihang Xiao, Yizheng Jiao, Jianheng Hou, Danyang Zhang, Pengcheng Xu, Boyang Zhong, Zehong Zhao, Gaoyun Fang, John Kitaoka, Yile Xu, Hua Xu, Kenton Blacutt, Tin Nguyen, Siyuan Song, Haoran Sun, Shaoyue Wen, Linyang He, Runming Wang, Yanzhi Wang, Mengyue Yang, Ziqiao Ma, Raphaรซl Milliรจre, Freda Shi, Nuno Vasconcelos, Daniel Khashabi, Alan Yuille, Yilun Du, Ziming Liu, Bo Li, Dahua Lin, Ziwei Liu, Vikash Kumar, Yijiang Li, Lei Yang, Zhongang Cai, Hokin Deng
show all authors
ICML, 2026
I contributed two tasks evaluating object counting and physical reasoning capabilities of video-language models.
Korean patent figure
Yong-Guk Kim, Soo-Mi Choi, Anh H. Vo, Tin Nguyen
Korean Patent No. 10-2979119 — Registered Jun 2026

Teaching

Academic Service

  • Reviewer for ARR Rolling Review (ACL, EMNLP, NAACL), UIST, HAI, Knowledge-based Systems