I'm Wei Xia (Xwell), a senior applied AI researcher and builder at WeChat, Tencent.

My current work focuses on AI applications for WeChat. Previously, I worked on coding models and agent evaluation for CodeBuddy and WorkBuddy. Before Tencent, I spent seven years at Huawei Noah's Ark Lab, working on large language models, reinforcement learning, recommendation systems, and personalized learning.

Two themes shape my work and writing:

• Solving practical problems with agents. I work across agent design, data, post-training, evaluation, and deployment, with an emphasis on reliable results and improvement through feedback from real use.
• Learning through practice. I value continuous learning through reading, writing, and building. Small experiments help me put ideas into practice, reflect on the results, and refine my understanding.

My current research interests include code post-training, training data selection, memory for long-horizon agents, and learning from user feedback. I also explore agent systems and model training for personalized content and engagement in social and marketing applications.

I hold a master's degree in Pattern Recognition and Intelligent Systems from the University of the Chinese Academy of Sciences and a bachelor's degree in Automation from the University of South China.

My interest in building began with control projects at university: a rotary inverted pendulum and a two-wheeled self-balancing car.