Presented at the ICML 2025 Workshop on Computer Use Agents: see video. Currently under revision for NeurIPS 2025.
An open-source AI assistant for STEM questions, aligned with Direct Preference Optimization and Supervised Fine-Tuning.
In AI for mathematics, the hardest step is translating informal problem statements into formal language. I built hypothesis decomposition and retrieval pipelines that improve LLM performance on this task.
4th prize @ Reprogramming AI Models Hackathon 2024. Unlearning harmful capabilities in LLMs while preserving general ones, in a cheaper and more interpretable way.
Collaboration with the European Space Agency. Machine learning models to detect solar events from spacecraft sensor data.
2nd prize @ AWS x Start Lausanne Hackathon 2025. AI forecasting solution and startup concept that predicts renewable energy oversupply with high granularity.
Data analysis of climate change content on YouTube. Explored video trends, interest surges, sentiment analysis, and channel influence to reveal how public engagement shifts over time.
1st prize @ AI Testing Hackathon 2022. Extended the Contrast Consistent Search method to test model latent knowledge under ambiguity, showing good generalization.
Looking to collaborate on a project? Need feedback or want to discuss ideas from my field? Exploring new opportunities or potential roles?
I'd be happy to collaborate, share insights, and exchange ideas: don't hesitate to reach out!