PhD student in Computer Science at Cornell University, advised by Claire Cardie, expected to graduate in 2026.
Background
Research Interests:
- Efficiency and Budget Aware Agent: How to design agentic frameworks that are efficient and adaptive under budget constraints (such as thinking budget, tool call budget, or interaction turn budget). Specifically, for math problems, it means reducing thinking tokens and preventing the model from overthinking. For deep research, it means given the budget of tool calls (such as search), teaching the model to use these tool call budgets wisely. For AI assistants, especially when user queries are ambiguous, it means asking the right questions and knowing when to stop asking for clarifying questions.
- Alignment and Safety: Interested in understanding, controlling LLM and agent behavior. As agentic systems become more powerful, human verification of the traces (which are usually very long) becomes infeasible. How can we verify that the agent reaches the answer through the correct path so that we can trust the answer?
- Using Agents to Accelerate Scientific Discovery: How can we build better agent scientists to help human scientists accelerate scientific discovery? How to verify that the scientific discoveries made by the agent are actually correct?