🌐 yixinhuang48.github.io — personal website 🔬 Hao AI Lab — research group 📚 Google Scholar — publications
Research Assistant at UCSD Hao AI Lab · M.S. Computer Science Student · San Diego, CA
I work on LLM systems, evaluation, and GPU-accelerated ML infrastructure.
"A journey of a thousand miles begins with a single step." — Confucius
- LLM evaluation & benchmarks (agents, games, scientific reasoning)
- Large-scale training & inference systems (FSDP, vLLM, Ray, Slurm)
- Reinforcement learning for agents (GRPO, NeMo-Gym)
| Project | Description |
|---|---|
| GamingAgent | LLM/VLM gaming agents for model evaluation — long-horizon reasoning, memory & perception (Doom, Sokoban, Tetris, Pokémon Red) |
| VideoScience | Benchmark for scientific correctness in text-to-video models — physics & chemistry concepts, VLM-as-Judge scoring |
| NVIDIA NeMo Gym | RL environments for LLM training — scalable RL, reward profiling, GRPO (Integrating Sokoban & Tetris) |
| lmenv | LLM environment framework for interactive evaluation — standardized interfaces for game-based agent testing |
Python · PyTorch · CUDA · vLLM · SGLang · NeMo RL · Ray · Docker · Slurm · FSDP · DeepSpeed · Linux · Git