I am a DDSA PhD Fellow at the Technical University of Denmark, affiliated with the University of Copenhagen and Rigshospitalet. My fellowship is funded by the Danish Data Science Academy through the Novo Nordisk Foundation and Villum Fonden. I am supervised by Prof. Lars Kai Hansen (Technical University of Denmark) and Prof. Sebastian Weichwald (University of Copenhagen).

Following a 2026 research visit to Stanford's STAIR Lab, where I worked with Prof. Sanmi Koyejo, I am now based at DTU in Copenhagen. My research develops causal and mechanistic methods for understanding, evaluating, and controlling learned representations, with recent work on reasoning-model hidden-state geometry, representation-level safety evaluation, steering, and trustworthy machine learning for clinical neuroscience.

Read More →

Based in Copenhagen, Denmark. Open to collaborations and new ideas.
Reach me at agjma [at] dtu.dk or gjoelbye [at] cs.stanford.edu.