About me

I am a PhD candidate in Artificial Intelligence at École de technologie supérieure (ÉTS) and a researcher at LIVIA in Montréal, Canada. My current research spans LLM agents, computer-use agents, agent evaluation, vision-language models, and vision transformers architecture.

Most recently, I was a Visiting Researcher at ServiceNow AI Research, where I worked on self-improving web agents that learn reusable skills, workflows, and memories from past interaction trajectories without fine-tuning. This work investigates a practical question: when do online memory and skill modules improve agent performance enough to justify their inference-token cost? I also contributed to AgentHorizon, a benchmark for evaluating agentic judges on long-horizon computer-use tasks.

My doctoral research also spans vision transformer architecture and pretraining, open-vocabulary semantic segmentation, few-shot adaptation of vision-language models, and generalized few-shot semantic segmentation.

Before my PhD, I earned an MSc in Artificial Intelligence and a BSc in Software Engineering from Sharif University of Technology. My master’s research explored disentangled representation learning with variational autoencoders. I also bring two years of production software-engineering experience from Divar, where I built and maintained backend services and web applications.