Yufeng Huang
AI systems engineer · Game AI · LLM-driven agents · Human-centered interactive AI
I am Yufeng Huang, an AI systems engineer with 8+ years of experience in game AI, NPC systems, interactive agent behavior, and AI-assisted production tools. I currently lead a small AI and tools group at Tencent SGRA Studio, where my work focuses on NPC intelligence, agent behavior, and production workflows for interactive virtual-world games.
My research interests center on human-centered AI agents for interactive virtual environments. I am especially interested in believable and controllable LLM-driven agents, personality and memory modeling, long-term human-agent interaction, and AI-assisted creative tools for narrative, quests, dialogue, and agent behavior.
Research interests
- Human-AI interaction for interactive agents and virtual companions
- LLM-driven agents with personality, memory, and long-term interaction
- Game AI, character AI, and autonomous agents in virtual environments
- AI-assisted authoring tools for narrative, quests, dialogue, and agent behavior
- Evaluation of expressed and behavioral personality alignment in LLM agents
Background
Before Tencent, I worked at ByteDance Nuverse and NetEase Games on gameplay and AI systems across strategy RPG, open-world ARPG, MMORPG, and MOBA projects. My engineering experience includes behavior trees, finite-state machines, utility AI, combat AI, ambient NPC behavior, dialogue and quest tooling, blueprint editors, narrative editors, and AI-assisted production pipelines.
I received my M.S. in Pattern Recognition and Intelligent Systems from Huazhong University of Science and Technology, where my thesis focused on CNN-based visual place recognition in changing environments.
Selected projects
- Mimosa: a runnable prototype for long-term multimodal virtual companions with Live2D, text chat, voice input, speech recognition, text-to-speech, LLM integration, personality modeling, and persistent memory.
- llm-persona-gap: a prototype framework for evaluating the gap between expressed personality and behavioral personality in LLM-driven agents.