Hande Dong
Head of Model R&D for Tencent WorkBuddy,
Head of Model R&D for Tencent WorkBuddy, graduated from the University of Science and Technology of China. He currently leads model research and development for the WorkBuddy team, focusing on establishing a data flywheel for models in WorkBuddy’s agentic AI scenarios and advancing toward continual learning. His areas of expertise include large language models, data flywheels, and production-oriented model engineering.
Topic
The Model–Data Flywheel for LLM Agents: Continuous Learning with WorkBuddy
As Agent products become increasingly widespread, more and more Agent experience data is being collected, capturing interactions among Agents, their environments, and users. Such data reflects the real-world scenarios Agents encounter, how they respond to those environments, and how users intervene during interactions. It contains rich, high-quality information grounded in real-world usage and user feedback, providing valuable signals for model learning. However, effectively transforming Agent experience data into high-quality training signals remains a significant challenge. WorkBuddy has long been focused on leveraging Agent experience data to improve model capabilities and building a data flywheel connecting users, data, and models. This approach enables continuous learning from real-world product experience and helps Agents improve over time. This talk will share WorkBuddy's perspectives and practical experience in building and leveraging the data flywheel, exploring how Agent experience data can be transformed into a sustainable source of model improvement.
