Research

My research interests center on efficient, trustworthy, and multimodal learning systems, with a current focus on medical AI and lightweight vision-language models.

Research Interests

• Efficient AI: model compression, efficient inference, lightweight architectures
• Trustworthy AI: robustness, fairness, interpretability
• Theoretical ML: information theory, learning theory
• Multimodal Learning: vision-language models, medical AI
• Reinforcement Learning: decision-making, sequential learning

Undergraduate Thesis

Development of a Multimodal Medical Assistance Chatbot for Domain-Specific Applications

Supervised by Prof. Mohammad Ariful Haque, with collaborators Tanvir Ahmed Khan and Ismam Nur Swapnil.

Developed a dermatology-focused multimodal medical assistant by fine-tuning a vision-language model on DermNet, using GRPO and DPO for reasoning and conversational alignment, integrating DINOv2 and knowledge-graph-based RAG for diagnostic precision, and applying structured pruning for efficient deployment.