Changxiu (Leo) Ji

I am a sophomore at Stanford University majoring in Mathematics and Computer Science. I am a member of the Stanford Pervasive Parallelism Lab and my adviser is Professor Kunle Olukotun. I have also been working with the Stanford Scaling Intelligence Lab and the Iris Lab.

My research work focuses on building AI systems that continuously learn from experience and improve over time (a topic known as recursive self-improvement). This includes: curriculum design (Learning what to learn, ICLR 2026 LLA); parallel execution frameworks (Combee, COLM 2026); understanding the failure modes of self-evolving agents; learning to recover from agent failures (Sentry, arXiv 2026); stateless language agents for long-horizon research (SLA, arXiv 2026). I am also an active contributor to many popular RSI frameworks like Agentic Context Engineering (ACE, ICLR 2026), GEPA (ICLR 2026), and CORAL (COLM 2026).

Besides research, I enjoy getting hands-on with real-world systems across different industries, from big tech and AI labs to quantitative finance. In the past, I was very fortunate to intern at Tencent WeChat Reading team (Long-context multi-agent systems), DeepSeek (DeepSeek sparse attention), Algovant (Option market agent), and Scientech Research (Agent swarm research for alpha discovery). Next summer, I’ll be joining TikTok as a software engineer intern in San Jose. If you’re around the area, I’d love to connect!

Before Stanford, I graduated from Shenzhen Middle School, where I won gold medals in national-level mathematics and programming olympiads. I am also deeply interested in pure mathematics, including the Hopf fibration, convex optimization, Galois theory, and chaotic maps with symbolic dynamics. I attended the Stanford Mathematics Camp (SUMaC) in 2023 and 2024, studying abstract algebra and algebraic topology.