My research interests lie in inference-time controllability and alignment, with a focus on methods that steer generative models toward safe and user-aligned outputs without requiring expensive retraining. Recently, I have been especially interested in safe generation and discrete diffusion.