Md Mehedi Hasan (মেহেদী)

Welcome.

I am Md Mehedi Hasan, an AI researcher working on problems at the boundary of machine learning security, systems-level trust, and real-time computing.

My research starts from a single conviction: AI systems deployed in critical environments must be simultaneously capable, secure, and formally analyzable. This drives work across several interconnected directions.

On the security side, I design adversarial defense mechanisms for large language models and hardware platforms. SentraGuard, my multilingual LLM jailbreak defense framework, combines transformer-based classification, FAISS-indexed semantic retrieval, and reinforcement-enabled human oversight to achieve real-time protection against complex prompt-level attacks. FAARM addresses the hardware layer, building firmware attestation for GPU Trusted Execution Environments against state-of-the-art MCU injection exploits.

On the systems side, ReDAGRT formalizes multi-DAG scheduling for ROS 2, bringing Rate-Monotonic schedulability theory to robotic middleware and enabling deterministic execution in safety-critical cyber-physical deployments.

In healthcare AI, CLIN-LLM integrates retrieval-augmented generation with uncertainty-aware clinical NLP, and my federated learning work on Alzheimer's detection demonstrates that diagnostic accuracy and patient privacy are not competing objectives.

Every project on this portfolio was executed independently, from initial idea through implementation, experimentation, and submission to top-tier IEEE and ACM venues.

I am currently applying to PhD programs in Computer Science. If you are a researcher working on trustworthy AI, LLM security, or real-time systems, I would be glad to connect.

Md Mehedi Hasan