Mohamad Bazzi

Software Engineer & Computer Science Professional

Mohamad Bazzi Mohamad Bazzi presenting

I am a passionate Senior Computer Science and Engineering student at the American University of Beirut with a strong foundation in artificial intelligence and machine learning. My research interests include large language models (LLMs), reinforcement learning from human feedback (RLHF), retrieval-augmented generation (RAG), active learning, and test-time fine-tuning. I am broadly interested in advancing AI capabilities and exploring their applications across diverse domains such as healthcare, education, and beyond.

Education

2022 - 2026

Bachelor of Engineering in Computer Science(CSE)

American University of Beirut

  • AI and data science concentration area
  • Minor Mathematics
  • President of the Financial Intelligence Club
  • Founding member and lecturer at AI Journey

Experience

Research Assistant

University of British Columbia

Mar 2025 – Sep 2025

Advised by Dr. Danica Sutherland and working closely with Dr. Wonhoe Bae, we explored active learning for preference alignment and test-time fine-tuning. My work focused on the use of herding models and Gaussian Process uncertainty measurements to enhance large language models' ability to retrieve relevant documents for user queries and adapt their performance dynamically across tasks.

Research and Development

Maroun Semaan Faculty of Engineering and Architecture

Jan 2025 – Present

Worked with Professor Ammar Mohanna to develop an agentic RAG-based advising system that scrapes all university resources to answer any student question—from course instructors to office locations. The project won the Best Project Award in the AI in Industry course and is now being extended as our FYP, set to launch for 1,000+ students.

Undergraduate Researcher

American University of Beirut

Sep 2024 – Dec 2024

Conducted research on Optimizing Deep Neural Networks via Safety-Guided Preservation Sets (arXiv preprint), under the supervision of Professor Ammar Mohanna. The project introduced a novel compression framework that enhances generalization and reduces variance through preservation-driven quantization. This work was recognized among the Top 10 papers in the IEEE Student Paper Competition.

Undergraduate Researcher

American University of Beirut

Jan 2025 – May 2025

Designed a multi-agent retrieval system where a master model routes queries to section-specific agents for targeted question answering. Developed a progressive graph-building mechanism that incrementally expands document knowledge graphs with structured answers and source references, improving multi-hop reasoning and document comprehension.