profile

Weixian Xu

I am an undergraduate in the ACM Honors Class at Shanghai Jiao Tong University, currently conducting research at Stanford SALT Lab with Prof. Diyi Yang. Previously, I worked at GAIR Lab with Prof. Pengfei Liu, and I have collaborated with Prof. Mengdi Wang and Shilong Liu at Princeton/AI2.

Research Interests

• Continual Self-Improvement: Agents that analyze their own behavior, reflect on outcomes, retain useful knowledge, and improve through long-horizon research and tool-use loops.
• Human-Centered Intelligence: Self-improving systems that become more capable while helping people learn, work, collaborate, and make better decisions.

Key Projects

• ASI-Evolve: Closed-loop self-improving research framework for AI agents.
• EEVEE: Multi-dataset test-time prompt learning for self-improving agents in real-world task streams.
• ASI-Arch: Autonomous neural architecture discovery system (1.1k+ GitHub stars).
• ACore: Comprehensive OS kernel implementation in Rust.
• Imxc: High-performance compiler with advanced optimizations.
• RISC-V CPU: Verilog implementation of Tomasulo architecture.

Background

My research is dedicated to building AI systems that can recursively self-improve toward broader and more human-beneficial intelligence. I view intelligence as the ability to learn from analysis and reflection, improve over time, and help people in meaningful ways.

My systems background in OS kernels, compilers, and hardware architecture helps me approach this work from both algorithmic and engineering perspectives. I aim to build AI systems that are not only conceptually interesting, but also implementable, testable, and able to improve through real feedback loops.

Contact

Email: hz.czar@sjtu.edu.cn

GitHub: @HZxCzar