Image of Taminul

Md Zabirul Islam (he/him)

islamm11 [at] rpi [dot] edu
2145, CBIS Building, 110 8th St, Troy, NY 12180, United States
| | | | |

Md Zabirul Islam is a Doctoral Student in Computer Science at School of Science, Rensselaer Polytechnic Institute, currently serving as a Graduate Research Assistant at the Wang-AXIS: AI-based X-ray Imaging System (AXIS) Research lab under the supervision of Dr. Ge Wang. His research focuses on Large Language Models (LLMs), Multimodal Foundation Models, and Generative AI, with an emphasis on medical imaging, structured clinical reasoning, and AI-enabled education systems. He develops structured, verifiable, and trustworthy AI for consequential knowledge workflows, spanning evidence-grounded generation, constrained decoding, retrieval, abstention and verification, and instructor-governed AI for education. His systems are deployed on institution-operated GPU infrastructure.

🚀

Internship Availability — Summer 2027

I am actively seeking internship opportunities for Summer 2027 in AI/ML Research. I am particularly interested in research roles focused on Large Language Models (LLMs), Multimodal Foundation Models, and Generative AI. I am eager to contribute to research-driven teams at leading AI labs, industry research groups, and innovative startups working on reliable, scalable, and deployable AI systems.

Biography

I build AI systems that people can check before they rely on them. Most of my work is on large language and vision-language models for two settings where a wrong answer costs something: clinical care and the classroom.

I started in Bangladesh. My undergraduate thesis at KUET used clustering to study public-sector corruption. Right after I graduated, COVID-19 hit, and I spent the lockdowns building deep learning models to detect COVID-19 from chest X-rays. One of those papers has been cited more than 800 times. From 2021 to 2023 I taught AI, image processing, algorithms, and numerical methods at Daffodil International University, to more than 1,000 students.


In 2024 I came to RPI to work with Prof. Ge Wang and moved to large language models. The question I keep coming back to is simple: when the model is wrong, can someone catch it? Each of my current projects answers it differently. In radiology, the model can only write values a clinical template allows, so free-text CT reports come out fully structured (F1 of 0.976 across two institutions). In clinical-trial onboarding, the system picks one trial or declines to answer, and explains only from that trial's evidence. MIP-BENCH tests whether vision-language models understand what a lecture slide is trying to teach, not just what is on it. ALIVE, now running in 48 lectures, lets an AI teach and quiz students while the instructor signs off on anything that changes the course. The clinical work is with collaborators at the NIH Clinical Center, Yale, MGH/Harvard, and Wake Forest.

My work has 1,254 citations (h-index 7, i10-index 6). I have been lucky to work with Prof. Ge Wang, Dr. Chuang Niu, Dr. Md. Milon Islam, Dr. Tianfan Fu, Prof. Weiping Ding, and Prof. M.M.A. Hashem. I also review for the Medical Physics journal.

Recent Publications

  • Md Zabirul Islam, Md Motaleb Hossen Manik, and Ge Wang. “CourseBlueprint: A Structured Pipeline for Adaptive Pedagogical Video Generation Grounded in Course Corpora.” In arXiv , 2026.
  • Md Zabirul Islam, Md Motaleb Hossen Manik, Oge Marques, Hisham Mohamed, and Ge Wang. “ALIVE: An Avatar-Lecture Interactive Video Engine with Content-Aware Retrieval for Real-Time Interaction.” Submitted to IEEE Transactions on Learning Technologies, 2026.
  • Md Motaleb Hossen Manik, Md Zabirul Islam, and Ge Wang. “MILU: A Consensus Ensemble Benchmark for Multimodal Medical Imaging Lecture Understanding.” In Journal of Medical Imaging , vol. 13, no. 6, 062202, 2026.
  • Md Motaleb Hossen Manik, Md Zabirul Islam, and Ge Wang. “MEDI-SLATE: Medical Imaging Slide-Lecture Aligned Teaching Ensemble.” In Visual Computing for Industry, Biomedicine, and Art , vol. 9, article 8, 2026.
  • Md Motaleb Hossen Manik, W. Muldowney, Md Zabirul Islam, and Ge Wang. “Development of an Optically Emulated Computed Tomography Scanner for College Education.” In Visual Computing for Industry, Biomedicine, and Art , 2026.
  • Md Motaleb Hossen Manik, Md Zabirul Islam, and Ge Wang. “SlideChain: Semantic Provenance for Lecture Understanding via Blockchain Registration.” In arXiv , 2025.
  • Md Motaleb Hossen Manik, Md Zabirul Islam, and Ge Wang. “N-ReLU: Zero-Mean Stochastic Extension of ReLU.” In arXiv , 2025.
  • Md Zabirul Islam and Ge Wang. “Avatars in the Educational Metaverse.” In Visual Computing for Industry, Biomedicine, and Art , 2025.
  • Md Sayed Tanveer†, Md Zabirul Islam†, Chuang Niu, Parisa Kaviani, Qing Lyu, Mannudeep K. Kalra, Christopher T. Whitlow, and Ge Wang. “Development and Validation of a Large Language Model for Generating Fully-Structured Lung Nodule Reports.” Submitted to Journal of the American Medical Informatics Association (JAMIA), 2026. (†equal contribution)
  • Ling Yue, Jonathan Li, Sixue Xing, Md Zabirul Islam, Bolun Xia, Jintai Chen, and Tianfan Fu. “TrialDura: Hierarchical Attention Transformer for Interpretable Clinical Trial Duration Prediction.” In NeurIPS 2024 Workshop on AI for New Drug Modalities , 2024.

Education

  • Doctor of Philosophy (PhD) in Computer Science, Spring 2024 - Present
    RPI Logo Rensselaer Polytechnic Institute, Troy, USA
  • Master of Science (MSc) in Computer Science, Fall 2025
    RPI Logo Rensselaer Polytechnic Institute, Troy, USA
  • BSc in Computer Science and Engineering, Mar 2020
    kuet Logo Khulna University of Engineering & Technology, Bangladesh

Experience

  • Graduate Research Assistant, June 2025 - Present
    RPI Logo Rensselaer Polytechnic Institute, Troy, USA
  • Graduate Teaching Assistant, January 2024 - May 2025
    RPI Logo Rensselaer Polytechnic Institute, Troy, USA
  • Lecturer, February 2021 - December 2023
    DIU Logo Daffodil International University, Bangladesh

News

June 2026 New preprint "CourseBlueprint: A Structured Pipeline for Adaptive Pedagogical Video Generation Grounded in Course Corpora" released on arXiv. arXiv
2026 Submitted "ALIVE: An Avatar-Lecture Interactive Video Engine with Content-Aware Retrieval for Real-Time Interaction" to IEEE Transactions on Learning Technologies.
2026 Our journal article "MILU: A Consensus Ensemble Benchmark for Multimodal Medical Imaging Lecture Understanding" has been published in Journal of Medical Imaging. Full Text
2026 Our journal article "MEDI-SLATE: Medical Imaging Slide-Lecture Aligned Teaching Ensemble" has been published in Visual Computing for Industry, Biomedicine, and Art. Full Text
2026 Submitted "Development and Validation of a Large Language Model for Generating Fully-Structured Lung Nodule Reports" (co-first author) to the Journal of the American Medical Informatics Association (JAMIA).
January 2026 Our journal article "Development of an Optically Emulated Computed Tomography Scanner for College Education" has been published in Visual Computing for Industry, Biomedicine, and Art. Full Text
December 2025 Successfully completed my M.Sc. degree in Computer Science at Rensselaer Polytechnic Institute.
December 2025 Presented a poster on my master’s research project, “ATOMIC: Avatar-based Teaching for an Optimized Medical Imaging Course,” supervised by Prof. Ge Wang. The project introduces an AI-driven, avatar-based educational system designed to enhance scalability, interactivity, and multilingual delivery in medical imaging education.
December 2025 New preprint "SlideChain: Semantic Provenance for Lecture Understanding via Blockchain Registration" released on arXiv. arXiv
December 2025 New preprint "ALIVE: An Avatar-Lecture Interactive Video Engine with Content-Aware Retrieval for Real-Time Interaction" released on arXiv. arXiv
November 2025 New preprint "N-ReLU: Zero-Mean Stochastic Extension of ReLU" released on arXiv. arXiv
October 2025 Served as a peer reviewer for a manuscript submitted to the Medical Physics journal.
June 2025 Our journal article "Avatars in the Educational Metaverse" has been published in Visual Computing for Industry, Biomedicine, and Art (vol. 8, no. 1).
February 2025 Our conference paper "Unifying Sentiment Analysis and Emotion Recognition for Bangla Text: A Hybrid Approach" has been presented at the 2025 International Conference on Electrical, Computer and Communication Engineering (ECCE), published by IEEE.