About
I'm a Game AI Engineer at Kuaishou GameMind Lab, where I lead a team focused on AI-driven game development, leveraging large language models to build intelligent tools and workflows that accelerate game creation. I also drive LLM post-training, including alignment and evaluation, to ensure our models are safe, reliable, and performant for real-world game scenarios.
I built arag-cli, a high-performance Agentic RAG search tool for AI agents, now open-sourced. Built in Rust, it provides keyword, semantic, and hybrid search with 50× speedup over Python alternatives.
Previously, I specialized in real-time video generation and discrete video tokenization at Soul AILab. I hold a Master's degree in Software Engineering from the University of Sydney, where my research on fine-grained visual categorization established the foundation for my expertise in visual computing and AI systems.
News
Jul 2026
🚀 arag-cli released as open-source: a high-performance RAG search tool for AI agents. 160ms keyword search, 5.1MB binary, 110MB memory. Built in Rust. GitHub
Jul 2026
🚀 TRACE-Bench released as open-source: Task-driven Roleplay Agentic Checklist Evaluation benchmark. GitHub · Project Page · Paper
Apr 2026
📄 Cutscene Agent released as Technical Report on arXiv. An LLM agent framework for automated end-to-end 3D cutscene generation, featuring a multi-agent system (director + specialist subagents for animation, cinematography, and sound design) with visual reasoning feedback, plus CutsceneBench for hierarchical evaluation. arXiv · Code
Dec 2025
🎉 RAP accepted to ICASSP 2025.
Mar 2025
🎉 Teller accepted to CVPR 2025.
Publications
Contact
Feel free to reach out for collaborations, discussions, or just to say hello.
zhangzw14@outlook.com