Here is Hongji Pu (JP, 蒲洪基).
An AI systems researcher driven by a fundamental curiosity: how can complex systems make better decisions under uncertainty?

My intellectual journey began with financial systems, where I studied how mathematical models capture uncertainty, quantify risk, and support decision-making in dynamic environments. Through this lens, I became fascinated by a broader question: as decision-making systems evolve, could they move beyond prediction and optimization to perceive, reason, learn, and continuously adapt?

My research focuses on language agents, continual learning and self-improvement, and world models for evaluation and decision-making. I study how language models and agents use memory and external evidence, learn from research findings and feedback, and acquire new capabilities while retaining what they have learned. I develop methods to verify proposed updates and determine when past experience remains useful as a system changes. I also investigate how world models guide rubric optimization for automated grading and algorithm design on complex networks. Across these directions, I ask when improvements in a model translate into better decisions.