Keshav Ramji

I currently research LLM reasoning, alignment, and self-improvement at IBM Research AI. 

 

Broadly, my research lies at the intersection of natural language processing and statistical machine learning. I aim to build a deeper empirical and mathematical understanding of foundation model behavior towards developing more capable AI agents. In particular, some areas I'm currently excited about are:

​

• Designing self-improvement algorithms and recipes; aligning foundation models at scale to human preferences

• Leveraging simple yet effective methods to improve reasoning in foundation models, across the model development pipeline (pre-training, reinforcement learning, leveraging test-time compute)

• Incorporating statistical techniques to build more robust and uncertainty-aware language modeling methods towards human-level decision-making

​

I graduated from the University of Pennsylvania in 2024, completing three degrees simultaneously, in Computer Science (M.S.E/B.S.E) from the School of Engineering, along with Statistics and Finance (B.S. in Economics) from the Wharton School. I have been fortunate to be advised by Prof. Surbhi Goel, Prof. Weijie Su, and Prof. Aaron Roth across various facets of language modeling research. I was also the founder and president of MLR@Penn, the first student-led research organization and community at Penn.​​​

​

I'm open to collaborations or talking about anything related to ML and research -- I especially enjoy mentoring younger students interested in getting involved in the field! Feel free to reach out at keshavsramji@gmail.com.​