I am a PhD student at the Max-Planck Institute for Intelligent Systems and the Ellis Institute of Tubingen under the supervision of Maksym Andriushchenko in the AI Safety and Alignment group. I completed my Masters in Artificial Intelligence at Université de Montréal, Mila and CRCHUM, co-advised by Prof. Bang Liu and Dr. Quoc Nguyen as part of the Applied Computational Linguistics Lab.

My current research interests are on understanding and evaluating the long-horizon capabilities of AI Agents, with particular emphasis on understanding and quantifying potential risks associated with increasingly autonomous AI systems. Currently, my main focus revolves around the task of forecasting which I believe to be an essential benchmark to properly assess AI capabilities. During my Masters, I mainly explored how large language models (LLMs) can be aligned and calibrated by leveraging insights from model interpretability with a focus on concept-based explanations. In brief, I am currently broadly interested in:

• Evaluation
• AI Safety/LLM Alignment
• Activation-based jailbreaks and defenses

During my undergraduate and graduate studies, I also led the UdeM AI undergraduate club to participate in many networking and conference events as well as participated in many volunteering initiatives. Outside of research, I am a big fan of racket sports.

Research

Activation Steering for Conditional Molecular Generation (AI4Mat Workshop @ NeurIPS 2025)

Using concept bottleneck models and activation steering, we enable conditional molecular generation by directly manipulating internal LLM representations. We show that our methods significantly improves LLM’s generation alignment to conditionned properties. Read more

Multilingual assessment of stereotypes in large language models (NAACL 2025)

We build and assess multilingual stereotypes across different LLMs. Read more

Calibrating Large Language Model’s with Concept Activation Vectors for Medical QA

We propose a novel framework for calibrating LLM’s uncertainty through Concept Activatoin Vectors. This improves the safety and calibration of LLMs in high-stakes medical decision making.
Read more

Atypicality-Aware Calibration of LLMs for Medical QA (EMNLP 2024 Findings)

We propose a novel method for elicitating LLM’s confidence in Medical QA by leveraging insights from medical atypical presentations. Read more