I am a Senior Research Scientist at Meta SuperIntelligence Labs (MSL), working in the core post-training team for Llama multimodal models, originally from California. My team was responsible for post-training and releasing the Llama 4 and Llama 3.2 multimodal (text + image) Model. I was responsible for improving performance in accuracy, reasoning, and chat, as well as developing post-training infra (SFT, DPO, RL, graders, Reward Models, evaluation).

Previously, I completed my PhD at Purdue University working with Professor Dan Goldwasser in Natural Language Processing and Multimodal understanding, originally from California. My PhD Dissertation was focused on Interactive Learning and Social Media Understanding. I published 5 first author papers (along with several workshop papers), and my work has been accepted at top NLP conferences like ACL, NAACL, EMNLP, and AACL.

I have interned at Microsoft Research, Amazon Science, and Fujitsu Laboratories, along with several Series A startups.

Broad Research Interests

I’m broadly interested in Natural Language Processing and Image Understanding, and improving AI in these areas. I’m interested in building an artificial agent that can interact with humans to comprehend Natural Language and use them to do better, especially without being re-trained, in a continuous online fashion.

I worked on this in my PhD Fake News Detection, Robotics, and Grounded Language Learning.

General Motivation: When AI agents make mistakes, they usually have to be re-trained. However, instead, humans should be able to help the agent, and the agent should be able to take advantage of this knowledge to do better in the future, especially in a similar situation. Enabling this would allow agents to continually improve when deployed in the wild, and be more interactive. Particularly with diffcult to train/access LLMs, this can be useful. Having strong LLMs that can understand and reason about the world is the first step towards this.