Nexgent
A desktop agent that works in your project folder, organizes its own team, and remembers what worked.
I am Wangjie Su, an undergraduate at Peking University pursuing a B.S. in Theoretical and Applied Mechanics and a double degree in Intelligent Science and Technology. My current work centers on reliable AI agents that can use domain tools, preserve state, recover from failures, and complete long-horizon scientific and industrial workflows.
I am currently studying long-horizon research autonomy: how agents can preserve executable state across interruptions, advance only through independent evidence, recover from failures, and accumulate reusable experience under explicit governance. I built Nexgent, a desktop agent whose coordinator splits work across background threads and improves how it divides work from measured outcomes, and which serves as the execution substrate for this research, including monitoring long fusion simulations; I also work on combustion modeling and general-purpose agent runtimes.
I am actively seeking internship opportunities in AI agents, agent harness and runtime infrastructure, and embodied intelligence—especially work connecting long-horizon agents with scientific workflows, multimodal perception, world models, or physical action.
谁终将声震人间,必长久深自缄默;谁终将亮如闪电,必长久如云漂泊。——尼采
to (expected)
Peking University
to (expected)
Peking University
GPA: 3.818/4.0. Full coursework is listed in my CV.
to
Hangzhou Yuanxi Technology
Contributed end to end across the frontend, backend, and runtime of ManySelves, independently designing and implementing its runtime kernel and capability packages.
to Present
AI + Fusion Energy Series Project, Xinao Technology Innovation Center, Peking University
PIs: Prof. Yunfeng Liang, Prof. Songfang Huang
Studying long-horizon research autonomy: preserving executable research state across interruptions and advancing only through independent evidence. The project develops four layers for orchestration/execution, evidence/verification, memory/evolution, and governance/collaboration; Nexgent provides the long-running substrate and can monitor long fusion simulations.
to Present
Combustion Laboratory, Peking University
PI: Prof. Zheng Chen
Studying multiscale combustion processes and the complementary roles of mechanistic and data-driven models.
A desktop agent that works in your project folder, organizes its own team, and remembers what worked.
A recoverable local runtime for file-defined agent teams, with a bundled power-distribution safety reporting capability.
A from-scratch CUDA framework, grown from course homework until it trains ImageNet-1k models.
Dig through soil to route water into the goal — a Position-Based Fluids puzzle game.
Where I have been, and where this site is read from. Click a highlighted country to zoom in, then a province for its cities.
Loading map…
Illustrative map. Boundaries from Natural Earth (China point of view) and DataV GeoAtlas / AMap; South China Sea islands and dashed line shown, Kashmir uncoloured.
Click the water to feed the fish · 点水面喂鱼