About Me

I am an Industry Professor of Computer Science and Engineering at New York University’s Tandon School of Engineering, and a newly minted PhD researching mechanistic interpretability and LLM security. I’m drawn to research that ships: building practical safeguards against unsafe and unintended model behavior, not stopping at the paper.

My dissertation, “Surgical Repair of Insecure Code Generation in LLMs,” diagnoses where a model goes wrong and patches it without retraining. So far I’ve applied the technique to attention-head bugs and prompt-injection defense. I completed my PhD in the ECE Department, advised by Siddharth Garg and Brendan Dolan-Gavitt, and I am a member of the Ensure Research Lab.

Before the PhD, I spent 15+ years at Microsoft as a principal engineering leader, working on Mobile, Office, and Hotmail. A lot of debugging transformer internals is the same instinct I used shipping production systems, just a layer deeper.

I have published in USENIX Security and ICML, and I teach graduate courses in deep learning, machine learning, distributed systems, and operating systems at NYU. I’m looking for industry roles and collaborators who want to build on this work, especially teams working on model safety and secure code generation.

Contact

You can reach me at my NYU zoom room.