Home
I’m a ELLIS PhD student at the Max Planck Institute for Intelligent Systems, advised by Prof. Moritz Hardt. My research is broadly in the domain of LLM post-training and AI evals.
These days, I am most excited about:
- synthetic data (problem) generation pipelines for RL.
- pushing model's capabilities in hard exploration problems (where pass@k is ~0) via curriculum learning techniques. The end-goal here would be to approach scientific discovery problems like open math conjectures.
- agentic training of LLMs for long-horizon tasks with tool use.
Past Work: My recent project has been on scaling data (synthetically) for training language models to become better at forecasting future events with RL. I also like working on meaningful evaluations and have pushed for shifting the QA evals ecosystem from MCQ to more open-ended benchmarks.
News
Old
- [June'25] I will be attending the YC AI Startup School from June 15-21. If you are in SF and would like to chat anything about startups or AI research (evals, scalable oversight, forecasting, and more), do reach out to me!
- [Sept 2024] Started PhD in Tübingen, Germany!
- From Feb. to May 2024, I was in Toronto working with Prof. Gillian Hadfield at the Vector Institute exploring the scope of normative alignment in RL-based agents.
- [Feb 22nd] Our work on fair sequential decision making won the Outstanding Paper Award at AAAI'24!
- From 21st June'22, I will be in Vienna, Austria, attending SoCS'22 and IJCAI'22.
- In May 2022, I began my research internship at LAMSADE, Université Paris Dauphine - PSL under Dr. Jérôme Lang and Dominik Peters. I am working at the intersection of computational social choice and automated decision-making, focusing on long-term fairness in the paradigm of virtual democracy.
Collaborate?
If you are working on a project and feel I could be worth collaborating or a challenging problem in which I might be interested, you can reach out to me here