Hamza Mooraj

ML Engineer and AI Researcher building and evaluating multimodal and agentic systems.

I’m an ML engineer and AI researcher interested in how multimodal and agentic systems behave, fail, and can be made more reliable in real-world settings. I graduated from Heriot-Watt University in 2025 with First-Class Honours in BSc Computer Science (Artificial Intelligence).

My work centres on building and evaluating AI systems under realistic constraints, especially where distribution shift, long-horizon interaction, retrieval, and deployment conditions affect behaviour. I focus on designing systems and evaluation frameworks that reflect real use, where failures can emerge and reliability becomes critical.

In my crop disease classification research, I developed AgriPath-LF16, a benchmark with explicit lab–field domain separation, and used it to evaluate architectural trade-offs across CNNs, contrastive vision-language models, and generative vision-language models. More broadly, I have built and analysed multimodal and retrieval-based systems across research, academic, and applied contexts.

My broader interests lie in evaluation, monitoring, and reliability for agentic and multimodal AI systems, especially in settings where systems must operate over time, interact with tools or other agents, and remain dependable under real-world complexity. I am particularly interested in understanding how and why these systems fail, and how such failures can be detected and mitigated.