AI for Scientific Discovery. I want to build agents that take part in the full research cycle—surveying the literature, proposing hypotheses, designing and running experiments, and interpreting the results. I am especially interested in how agents can judge which ideas are worth pursuing, and how to evaluate whether their work genuinely advances a field rather than merely looking plausible.

Self-Evolving Agents. I am interested in agents that keep improving after deployment by learning from feedback, accumulating memory and skills, and refining their own strategies, rather than staying fixed once trained. I care about making this improvement reliable as tasks and environments shift. Scientific research is a natural testbed: every experiment yields feedback an agent can learn from.