I am a research scientist at Meta, where I work on LLM post-training and evaluations for safety. My research focuses on developing methods that improve safe and reliable outcomes for large language models, including post-training techniques, synthetic and hybrid data design and generation, and model evaluation.
More broadly, I am interested in multimodal representation learning (vision and audio), LLM post-training, and optimization. Previously, I have worked on signal processing, compressed sensing, and randomized linear algebra. A selection of my recent research can be found here and the full list of publications can be found on Google scholar.
I obtained my Ph.D. in Electrical and Computer Engineering from Georgia Institute of Technology in 2020, where I was advised by Dr. Justin Romberg and worked closely with Dr. Mark Davenport. My Ph.D. research focused on subspace learning and sketching methods for high-dimensional linear algebra, with applications in array signal processing.