Hello! I’m Truong-Binh Duong, a Computer Science graduate from the University of Science – VNU-HCM, specializing in multi-modal AI and Vision-Language Models.

I recently completed the university’s High-Quality Program with an exceptional GPA of 3.98/4.0, earning Dean’s List recognition for 5 consecutive semesters. My graduation thesis, Counterfactual Reasoning for Robust Visual Question Answering, explored counterfactual training strategies that boost out-of-distribution robustness without extra annotations.

🎯 Current Focus

I’m pursuing AI Research Intern roles where I can advance state-of-the-art multi-modal reasoning systems. I enjoy rigorous experimentation, benchmarking, and translating research insights into production-ready models using PyTorch, Transformers, and modern MLOps practices.

🔬 Research & Publications

I have 2 peer-reviewed publications spanning Vietnamese Visual Question Answering and text-rich image analysis:

• ViVQA-X (ICISN 2026, LNNS by Springer Nature) — introduced the first Vietnamese VQA dataset with natural language explanations via an automated, multi-LLM pipeline, now officially published with DOI 10.1007/978-981-95-1746-6_18.
• Describe Anything Model for VQA on Text-rich Images (ICCV Workshop 2025) — led inference and benchmarking of Vision-Language Models across six datasets.

💼 Professional Experience

Currently serving as a Teaching Assistant at AI VIETNAM, where I design curriculum, author assessments, and produce teaching assets covering Machine Learning, Deep Learning, AI Agents, and LLM Reasoning.

🚀 Technical Expertise

• Programming: Python, C/C++, SQL
• Data Science: Pandas, NumPy, Matplotlib, Seaborn, Scikit-learn, XGBoost
• Deep Learning: PyTorch, Hugging Face, Transformers, CNN, RNN, LSTM, MLP
• MLOps & Tools: Docker, MLflow, Weights & Biases, FastAPI, Streamlit, Google Cloud (Vertex AI)
• Languages: Vietnamese (Native), English (Professional Working Proficiency – VSTEP B2)

🛠️ Featured Projects

• Counterfactual Reasoning for Robust Visual Question Answering (Thesis, 2025): Developed batch-contrastive losses, counterfactual sample synthesis, and a 3-stage curriculum that achieved state-of-the-art performance on VQA-CP v2 for annotation-free methods. Read the project summary →
• Heineken Image Analysis Tool (Hackathon, 2024): Engineered a multi-model AI pipeline (YOLOv10, Owlv2, PaddleOCR, CLIP) with a FastAPI backend to automate brand compliance and safety audits from images.

I’m passionate about turning research into practical solutions and have hands-on experience with the complete AI development pipeline—from data collection and model training to deployment and scaling.

Feel free to reach out through the social channels below to discuss research opportunities, collaborations, or potential internships!