I'm an AI researcher working on Mechanistic Interpretability and AI Safety. Currently finishing my MSc in Artificial Intelligence at the University of Groningen, where my work spans interpretable alignment, sparse autoencoders, reinforcement learning, and world models. I'm a first author on several workshop papers at NeurIPS, ICLR, and NLDL.
- The Anatomy of Alignment: Decomposing Preference Optimization by Steering Sparse Features — Spotlight, NeurIPS Mechanistic Interpretability Workshop 2025. Introduced FSRL, a transparent alignment framework that steers models via interpretable SAE features.
- Self-Ablating Transformers: More Interpretability, Less Sparsity — ICLR 2025 Building Trust Workshop. A transformer architecture that dynamically ablates neurons and attention units during training to yield more interpretable models.
- World Model Agents with Change-Based Intrinsic Motivation — Oral, NLDL 2025. Adapted change-based intrinsic rewards to the DreamerV3 world model algorithm for sparse-reward settings.
📚 More work can be found on my website and Google Scholar.
- Research Fellow @ Apart Research (2024–2026)
- Research Mentee @ EleutherAI — Summer of AI Research (2025)
- Technical Alignment Lead @ Safe AI Netherlands
Outside of research, I'm an avid gamer and an enthusiastic game developer. I'm currently experimenting with building games to explore new ideas. You can find my latest game here.