I’m a final-year PhD candidate at ÉTS Montréal. I work on LLM agents and how to evaluate them, and on vision and vision-language models. I have first-author papers at EMNLP, ICLR, CVPR, and WACV,

Most recently, I was a Visiting Researcher at ServiceNow AI Research, where I studied self-improving web agents and contributed to a benchmark for judging long-horizon computer-use agents. I have also spent two years as a production software engineer before starting research.

Research

LLM agents and agent evaluation

Agents that learn reusable skills and memories from their own past trajectories can improve without fine-tuning, but those modules cost tokens on every task. I study when they are actually worth it, and how to evaluate agents and LLM judges so that comparisons hold up.

Vision transformers and dense prediction

How to make vision transformers trained for classification produce better spatial features for segmentation, without changing how they are trained.

Open-vocabulary and few-shot learning with vision-language models

Adapting models like CLIP to segment categories they never saw during training, and to new tasks from only a few labeled examples.

See my publications or my Google Scholar profile.