Tingyu.

01

GPU hackathon project

1st Place: Built an LLM-powered CUDA codegen system for PyTorch modules, enabling custom kernel generation and benchmarking on NVIDIA GPUs.

  • PyTorch,
  • CUDA,
  • DeepSeek/Qwen,
  • GRPO,
  • Flask
CUDA Codegen from PyTorch Modules (1st Place in Strong Compute GPU Hackathon)
Compllama (2nd Place in Llama Impact Hackathon)
AI-Powered Conversation System
Lane Simulation in Autonomous Driving
AI Directory Website Scraper