About me
I’m a computer science PhD student at UCLA, advised by Rafail Ostrovsky. My research focuses on cryptography and secure computation, particularly garbled circuits. Previously, I was a research engineer at Alpen Labs.
I’m also exploring AI safety and mechanistic interpretability, with a particular interest in how language models reason and use information from context.
Email · Google Scholar · GitHub · LinkedIn
Interpretability project
Investigating Attention Heads in In-Context Retrieval
This exploratory investigation builds on Mixing Mechanisms. I studied how changing the outputs of two attention heads in Qwen2.5-3B-Instruct affects retrieval of information supplied in a prompt.
- Research write-up: experiments, results and limitations.
- NNsight notebook: self-contained code, saved GPU outputs and interactive attention visualizations.
- Process note: why I chose the project, how I used AI and what I learned.
Selected publications
Duty-Free Bits: Projectivizing Garbling Schemes
Nakul Khambhati, Anwesh Bhattacharya, and David Heath.
ACM CCS 2026 · Paper · Code
Authenticated Garbling with Tensor Gates
Nakul Khambhati, Turan Vural, David Heath, and Rafail Ostrovsky.
ACM CCS 2026 · Paper · Code
Garbling Groth16 with Native Group Operations
Nakul Khambhati, Aaron Feickert, Christian Lewe, and Mukesh Tiwari.
Preprint, 2026 · Paper · Code
Learning notes
I also keep working notes on machine learning and AI safety as I learn. These range from technical walkthroughs to reflections on papers and blog posts.