About me

I’m an Applied Scientist III (Senior) at Amazon AGI, where I’m the technical lead for multi-stage LLM post-training recipes (SFT, DPO, and RL) and the evaluation work around them. I completed my Ph.D. in Electrical Engineering at the Signal Analysis and Interpretation Lab (SAIL) at the University of Southern California (USC), advised by Shri Narayanan.

My research connects machine learning and human behavior:

• Speech-language models. I co-developed SpeechVerse, a multi-task audio-language model, first-authored zero-resource speech translation and recognition with LLMs, and worked on the adversarial robustness of speech LLMs (SpeechGuard).
• Learning from subjective annotations. I developed a method that turns annotators’ comparisons into a single label using triplet embeddings, which won the AVEC 2018 gold-standard emotion sub-challenge. The code is open source: TripletEmbeddings.jl.
• Health and behavior from wearable sensors. I designed and ran the TILES studies of hospital workers and ICU residents with off-the-shelf sensors. Other labs have reused the open data sets.

I also review papers for journals and conferences in speech, affective computing, and machine learning. You can find my work on the publications page, and a short version of my CV.

Outside of work, I enjoy backpacking, cycling, skiing, and photography.

Contact: email (turn on JavaScript to see the address)