Ph.D. Candidate · LLM Post-Training & Reinforcement Learning
Institute of Automation, Chinese Academy of Sciences
The Key Laboratory of Cognition and Decision Intelligence for Complex Systems
🎯 Expected graduation: June 2027 — seeking full-time roles in LLM post-training / RL / agents.
I work on making large models reason and act better through post-training. My research spans reinforcement learning for LLMs, RLVR, process-level verification, and LLM agents — with a bias toward ideas that survive contact with production-scale training.
Currently focused on RL post-training at scale — building unified reward systems, multi-teacher on-policy distillation, and data recycling pipelines for large-scale LLMs.
Most recently a research intern in the RLHF team at Qwen (Alibaba), led by Pengyu Cheng, working on agentic & RL training of LLMs — contrastive RLVR, LLM self-play, and multi-teacher OPD. Previously at Zhipu AI, building real-user function-calling benchmarks and agent post-training with VERL.
- LLM Post-Training — RLHF / RLVR, reward modeling, policy optimization
- Reinforcement Learning — GRPO-family algorithms, sample efficiency, multi-task reward design
- LLM Agents — tool use / function calling, memory, planning, multi-agent games
- Test-Time Reasoning — process verification, verifier-guided sampling
Post-Training
LLM Agents
Frameworks
Languages & Tools
