Orion Reblitz-Richardson

Independent alignment researcher · Distiller Labs

I build tools and methods for embedding alignment into language models during pretraining, rather than patching it in afterward. My current research focuses on training-time monitoring and active steering of model representations, tracking how concepts form and evolve as models learn, and intervening when they drift.

Previously, I spent nearly a decade at Meta where I led interpretability research and PyTorch infrastructure teams. I co-authored Captum, one of the most cited interpretability libraries in the PyTorch ecosystem, and contributed to the launch of our PyTorch Foundation with the Linux Foundation. I also worked on ExecuTorch, Meta's accelerated on-device AI inference infrastructure.

I hold an M.Eng. and B.S. in Electrical Engineering & Computer Science from MIT.