RatulHasan

1. Research Assistant

   October 2026 - Present

   • Working on a team journal project mapping attacker-controllable IoT metadata as an indirect prompt-injection surface for LLM home agents.
   • Measuring which Home Assistant device and integration text fields reach its built-in LLM conversation agents, and how often injected instructions trigger unauthorised physical actions or leak private state in a simulated home.
   • Evaluating how provenance tagging, actuating-tool gating and action confirmation reduce attack success, and their cost in benign-task utility.
2. Research Intern

   June 2026 - September 2026

   • First-authored a systematic review of 82 studies on LLMs in mental health, appraising methodological rigour and risk reporting (hallucination, bias, safety, human evaluation) with a 24-field extraction instrument.
   • Studied catastrophic forgetting in federated continual learning for mental health text classification, separating label-space shift from domain shift on Reddit stress and depression data.
   • Benchmarked seven training regimes (FedAvg, FedProx, EWC, replay, centralized and joint baselines) across 10 clients and three seeds with a MiniLM encoder.

BAUET Computer Society, Dept. of CSE, BAUET

• Conducted programming classes for undergraduate students covering C, C++ and basic data structures.
• Led coding exercises and programming contest preparation sessions to strengthen students' problem-solving and analytical skills.
• Explained programming concepts to students of varying skill levels and supported the development of their programming fundamentals.

Lab AR

• Led a team project to develop a web-based Property Management System.
• Contributed to backend development using Laravel (PHP) and collaborated on feature implementation.