Three PhD students of computer science have been named Amazon Doctoral Fellows by the JHU + Amazon Initiative for Interactive Artificial Intelligence, or AI2AI.
Niyati Bafna, Krithika Ramesh, and Zheyuan “Brian” Zhang were selected as 2026-2027 fellows based on their outstanding publication records, research proposals, and mentor support.
Fellows receive a full stipend, 20% tuition, and student health insurance for the fall and spring semesters.
Meet the Fellows
Niyati Bafna focuses on evaluating and improving multilinguality in large language models; her recent work explores theoretical bounds for the curse of multilinguality. Previously, she worked on diagnosing generation failures for mid-resource languages, robustness methods for dialectal variation, in-context learning for low-resource languages, and massively multilingual benchmarking. Before coming to Johns Hopkins University, Bafna was a research engineer at Inria Paris and earned an Erasmus Mundus MSc in Prague and Saarbrücken. In her free time, she enjoys playing tennis, swing dancing, eating, getting people into Baba is You, and being undecided about which language she is learning. She is advised by David Yarowsky.
Krithika Ramesh, a doctoral student in the Center for Language and Speech Processing, focuses on privacy and memorization in language models and enabling the development of trustworthy systems while safeguarding individual privacy. Her current research spans the development of privacy-preserving methods that minimize the privacy-utility tradeoff, building evaluations to verify if privacy guarantees actually hold in practice, and assessing inference-time privacy risks that agents pose. Outside her research, she plays Baba Is You and does (unsolicited) outreach work trying to convince people to switch to Signal and ProtonMail. Ramesh completed her bachelor’s degree in engineering in India and is advised by Anjalie Field.
Zheyuan “Brian” Zhang combines AI and cognitive science with a focus on embodied AI and continual learning. His current research centers on developing latent world models and persistent memory for test-time continual learning agents that can effectively explore, acquire new world knowledge and skills, retain episodic memories, and plan over long horizons throughout their lifetimes without catastrophic forgetting. Before joining JHU, Zhang earned his MS in robotics and a graduate certificate in cognitive science from the University of Michigan, and a BS in computer science from the University of Massachusetts Amherst. Outside of research, he enjoys music, movies, and games. He is advised by Tianmin Shu and Daniel Khashabi.