About Me

I am a PhD student in Computer Science at Rensselaer Polytechnic Institute, advised by Dr. Mohammad Mohammadi Amiri. My research focuses on trustworthy and privacy-preserving AI, with emphasis on machine unlearning for Large scale language and vision models, multi-agent LLM safety, representation-level privacy control, and alignment.

I design methods that make large AI systems more reliable in high-stakes settings by combining theoretical foundations with practical evaluation. Recent projects include optimization-driven machine unlearning, information-theoretic privacy control for sequential multi-agent LLM pipelines, and benchmark development for harmful or manipulative model behavior.

Alongside academic research, I collaborate with industry teams to translate research into deployable systems. At IBM Research, I am working on secure KV-cache sharing mechanisms for multi-agent LLM architectures through the RPI–IBM Future of Computing Research Collaboration.

Before starting my PhD, I worked across AI engineering and quantitative systems as a technical lead and founding engineer, building large-scale ML and backend platforms in production environments.

I hold a B.Sc. in Electrical Engineering with a minor in Computer Science from the National University of Sciences and Technology (NUST).

Research Focus

• Machine unlearning and selective forgetting for foundation models
• Privacy and leakage control in sequential multi-agent LLM pipelines
• Alignment-preserving fine-tuning and safety-retention under adaptation
• Robust evaluation of harmful/manipulative model behavior
• Privacy-aware data valuation and trustworthy AI systems

Publications

• Information-Theoretic Privacy Control for Sequential Multi-Agent LLM Systems — Under review

• DarkPatterns-LLM: A Benchmark for Detecting Manipulative and Harmful Behaviors in LLMs — Under review

• OFMU: Optimization-Driven Framework for Machine Unlearning — International Conference on Learning Representations (ICLR 2026)

• Generative AI Agents for Controllable and Protected Content Creation — NeurIPS 2025 Workshop on Generative AI for Protected and Controllable Content (GenProCC)

Talks

• ML Theory Seminar: A New Approach to Machine Unlearning — ML Theory Seminar

Academic Services

• Teaching Assistant, Rensselaer Polytechnic Institute (RPI)
  • Introduction to Artificial Intelligence
  • Introduction to Algorithms
  • Computer Architecture and Operating Systems
  • Computer Organization
• Conference Reviewer
  • IEEE Conference on AI (CAI), 2026
  • AAAI, 2025

Visitor Map

Visitor map is currently being initialized.