My research focuses on mechanistic interpretability. Recently, I analyzed cross-attention OV circuits in text-to-image diffusion models and tested how changes to their spectral components affect generated images.

I’m interested in building on this work to study AI alignment and safety.