MD Azizul Hoque
Master of Research Candidate
School of Computing, Faculty of Science and Engineering, Macquarie University
"Researching how to make large language models safer, more aligned, and accessible across low-resource languages"
Master of Research Candidate
School of Computing, Faculty of Science and Engineering, Macquarie University
"Researching how to make large language models safer, more aligned, and accessible across low-resource languages"
I am MD Azizul Hoque, a Master of Research candidate at the School of Computing, Faculty of Science and Engineering, Macquarie University, working under the supervision of Dr Usman Naseem and Professor Mark Dras.
My research focuses on LLM Alignment, Low-Resource NLP, and AI for Social Good. I am passionate about ensuring that large language models are not only powerful but also safe, equitable, and accessible — particularly for communities and languages that remain underrepresented in mainstream AI development.
I hold a Graduate Diploma of Research & a Bachelor of Information Technology (majoring in Artificial Intelligence) from Macquarie University and a Bachelor of Science in Computer Science and Engineering (Transferred) from the International Islamic University Chittagong. These foundations in both applied AI and core computer science drive my current research trajectory.
My research sits at the intersection of language model safety, multilingual NLP, and socially responsible AI. I investigate methods to align large language models with human values while extending their capabilities to low-resource languages, aiming to make AI technologies safer and more inclusive for diverse global communities.
Exploring techniques to ensure large language models behave in accordance with human intentions and ethical principles, focusing on safety and reliability.
Developing and evaluating NLP methods for languages with limited digital resources (Bangla), addressing data scarcity through transfer learning and cross-lingual approaches.
Applying artificial intelligence to address societal challenges, ensuring technology serves underrepresented communities and promotes equitable access to information.
Will be updated soon
Selected peer-reviewed publications and works in progress.
Kulahara, M., , et al. (2025). A CNN-Based Framework for Geometric Alignment of Historical and Satellite Imagery. IEEE Access, 13, 132259–132268. https://doi.org/10.1109/access.2025.3589497
Govindarajan, V., , et al. (2025). MAGIC-Enhanced Keyword Prompting for Zero-Shot Audio Captioning with CLIP Models. In Proceedings of the 26th International Conference on Web Information Systems Engineering (WISE 2025), CORE-B.
Kashyap, G. S., , Srivastava, K., Jalan, P., Kar, S., & Naseem, U. (2025). How Far Can Current Tokenizers Go in Bangla? A Benchmark and a Morphology-Aware Solution.
Kashyap, G. S., , Dras, M., & Naseem, U. (2025). The Right Tokens for the Right Languages? Evaluating Tokenization Algorithms For Non-Latin & Low-Resource Languages.
I am always open to research collaborations in LLM Alignment, Low-Resource NLP, and AI for Social Good. If your work intersects with these areas and you're interested in exploring joint research opportunities, I would love to hear from you.
Usman Naseem — Lecturer (NLP), School of Computing, Macquarie University
Mark Dras — Professor (NLP), School of Computing, Macquarie University
Simona Frenda — Research Associate, School of Mathematical & Computer Sciences, Heriot-Watt University
Marco A. Stranisci — Computer Science, Università degli Studi di Torino
Gautam Siddharth Kashyap — Phd Scholar, School of Computing, Macquarie University
Yiran Zhang — Phd Scholar, University of Western Australia
Baizid Kamruzzaman — Graduate, Computer Science & Engineering, International Islamic University Chittagong
SocialNLP Lab — Macquarie University
Selected data science and machine learning projects.
A collection of data science projects showcasing exploratory analysis, modelling, and visualization skills.
Developed a multi-class image classification system using custom CNN architectures, hyperparameter tuning, and transfer learning with MobileNet. Includes comparative evaluation to identify the most effective model.
Built a credit risk classification system using Logistic Regression, KNN, Random Forest, and Multi-Layer Perceptron to predict loan default, with comparative model evaluation.
More projects will be added here.
More projects will be added here.
Feel free to reach out for research discussions, collaboration, or any questions.
