I'm Yuzhe, a software engineer and MSc Computing (AI & ML) student at Imperial College London. I build across the stack, from sandboxed runtimes and coding-agent infrastructure to compilers and user-facing AI products.
My current research focuses on secure code execution and runtime isolation for serverless and agent workloads. Outside research, I like turning ideas into things people can actually use.
| Product | What I built | Status |
|---|---|---|
| Zao · 灶 | A context-aware iOS cooking companion that brings recipes, pantry, meal plans, and hands-free cooking into one flow. | Launching soon |
| Kotodama | A multilingual AI narrative-learning app where stories remember your choices and turn dialogue, vocabulary, and grammar into a connected learning path. | Web live · iOS coming soon |
| Project | What it does | Stack |
|---|---|---|
| Agent Workspace | A workspace-local command, package, and event runtime shared by coding agents and humans. | Go |
| SourcePin | Pins UI feedback to durable DOM and source locations, then keeps final acceptance with the human. | TypeScript, Vite, MCP |
| Project | What it does | Stack |
|---|---|---|
| shimmy-wasm | Runs untrusted code in a capability-constrained WASM/WASI sandbox for serverless environments. | Python, WebAssembly, WASI |
| RedScript | A typed language, compiler, LSP, and toolchain that emits vanilla Minecraft datapacks. | TypeScript, compilers |
| @faster-crud | An end-to-end type-safe CRUD toolkit spanning 18 backend, ORM, and frontend packages. | TypeScript, NestJS, React, Vue, Svelte |
| Project | What it does | Links |
|---|---|---|
| Leverage OJ | A full-stack online judge with competitive ranking, AI-assisted game design, and sandboxed execution. | backend · judge |
| VibeCheck | Analyzes GitHub commit histories for AI-assisted vibe coding, with user and repository scores, leaderboards, and badges. | git-vibe.pages.dev |
Also used across current projects: SwiftUI · Hono · Fastify · Drizzle ORM · TypeORM · MCP · CodeMirror · VS Code Extension API · OpenGL
Current: serverless sandboxing, secure code execution, and runtime isolation for agent workloads at Imperial College London.
Previous work includes graph neural networks, efficient ML, and AI watermark removal.
Cooking and baking, photography, Minecraft, travel, and learning Japanese when I'm away from the terminal.



