Research
I am a research intern working with Prof. Ruoxi Jia at Virginia Tech. My research interests span efficient and reliable agent systems and models with stronger capacities for abstraction and reasoning.
I study how system design can make existing model capabilities more useful, and how model architectures and training signals can develop new capabilities. My recent work focuses on agentic optimization and LLM post-training.
I am applying to Computer Science PhD programs for Fall 2027.
Papers
- 2026.09Belief-Calibrated Optimization: An Explicit World Model for Agentic OptimizationA persistent, revisable world model guides optimization of the scaffold around a frozen LLM.
- 2026.03Diagnosing and Repairing Citation Failures in Generative Engine OptimizationFailure diagnosis guides targeted repairs to improve citation in generative search engines.
- 2025.12Rethinking Expert Trajectory Utilization in LLM Post-training for Mathematical ReasoningStudies how expert trajectory scale and difficulty affect supervised fine-tuning and subsequent reinforcement learning.
- 2025.06Do Thinking Tokens Help or Trap? Towards More Efficient Large Reasoning ModelDual Policy Preference Optimization regulates thinking-token behavior for more efficient reasoning.
Blog