Undefined-OS

Undefined-OS

First Prize Entry of OS Kernel Implementation Track, CSCC 2025 — National Finals

2025.3 – 2025.8 · Project Report

Undefined-OS is a Linux-like, componentized kernel written in Rust based on the Arceos component library. It supports thread safety across multiple CPU cores and can run on x86_64, riscv64, aarch64, and loongarch64, as well as QEMU and hardware boards. We provide a Linux-compatible system call implementation. The OS already runs Git, GCC, Vim, Sqlite3, Lua, Rust, and Python — you can develop an OS on our OS.

TREA SOLO Hackathon 2025

Face Sculpting Master

Outstanding Work Award, ByteDance Trae SOLO Hackathon 2025 · Trae's official repost

2025.8

A face-sculpting Agent built with Trae IDE. It generates 3D virtual avatars from user descriptions, produces corresponding Blender code, and supports lighting and environmental visual adjustments.

Gravity Connect Four

Gravity Connect Four

Reinforcement Learning

2023.05 – 2023.07

Implemented UCT (UCB + Monte Carlo Tree Search) for Gravity Connect Four. The AI reached a 99.9% win rate against human opponents, and 97 wins in 100 games against other chess algorithms.

NYU Teaching Assistant

NYU Teaching Assistant

2024.10 – 2025.01

Teaching assistant in the CS Dept of NYU Shanghai for one semester — practicing English communication in CST and experiencing an English-language university classroom.

Human Resource Machine

Human Resource Machine

2022.11 – 2023.1

Led a team to reimplement Tomorrow Corporation’s programming puzzle game Human Resource Machine in C++. Players write simple code to control a robot, training programming thinking without a specific language. The game also emphasizes code optimization.

RISC-V pipelined CPU

RISC-V Pipelined CPU

2024.11 – 2024.12

Designed a five-stage RISC-V pipelined CPU in SystemVerilog on Vivado, with interrupts and exception handling, MMU virtual address translation, instruction and data caches, and a VGA controller for image display.

MiniGLM

MiniGLM

Natural Language Processing

2023.9 – 2023.10

A MiniGPT-style transformer pretrained and fine-tuned on Q/A pairs from Jin Yong’s wuxia novels, so the model can answer questions about their themes and characters.