- 🎓 M.S. student in Intelligent Information Systems (MIIS) at CMU LTI (Aug 2025 - Dec 2026)
- 💡 Dual B.S. from UIUC (Computer Engineering) and ZJU (Electrical & Computer Engineering)
- 🧠 Passionate about LLMs, long-context reasoning, and reinforcement learning
- 🛠️ Previously interned at Microsoft Research Asia (MSRA) — worked on LongRoPE2 (ICML'25) and LoongRL (ICLR'26 Oral) for long-context LLM reasoning
- 🚀 Currently interning at NVIDIA on AI-agent tooling: a pluggable check API + Perforce backend for agent guardrails, agent-workflow skills benchmarked at up to -38% cost with no accuracy loss, and telemetry hardening for cost governance
- 🔊 Contributing to sglang-omni: native TTS model support + serving optimizations (torch.compile, CUDA Graph, LRU caching) cutting decode latency 5.5x
- 💼 Open to LLM-related Machine Learning Engineer / Research Scientist roles — graduating December 2026
- 📫 Reach me: gaokaiz@andrew.cmu.edu — 📄 Resume
- 🧑💻 Hybrid-Gym — Synthetic task generation for training coding agents to generalize across repository-level environments (ICML 2026)
- 🧾 LongRoPE2 — Extended LLM context to 128K tokens with >98.5% short-context retention (ICML 2025)
- 🚀 LoongRL — RL framework enabling 7B models to outperform 32B LRMs on 100k-200k token reasoning (ICLR 2026 Oral)
- 🐒 Stochastic Monkeys — Robustness benchmarking of LLM safety alignment
