Recently, the LongCat Interaction team of Meituan released a technical report on a large model interaction system called "WOWService," which aims to address multiple challenges faced by large model technology in the local life service sector.

The report points out that there is a mismatch between industry-wide general capabilities and domain-specific needs. It is difficult to balance the reliability and personalization of services in complex scenarios, and high data costs and long training cycles further increase the difficulty of development. In addition, the industry lacks reusable business adaptation frameworks and real-scenario optimization solutions, leading to low efficiency in technology implementation.

image.png

The WOWService system significantly enhances the reasoning ability and professionalism of intelligent customer service by integrating cutting-edge technologies such as multi-agent collaboration, reinforcement learning, and domain knowledge enhancement. The system uses human-computer collaborative annotation, model self-criticism reinforcement, and knowledge rewriting technologies, demonstrating stronger flexibility in handling complex instructions and multi-task scenarios. It can achieve the same effect as traditional solutions with only 10% of the small model annotation data, thereby reducing training costs and time.

The WOWService system has been widely applied in Meituan's intelligent customer service system, covering dozens of business scenarios, building high-quality multi-turn dialogue data, and improving the data construction system. Through continuous system optimization, this project not only improves user satisfaction but also surpasses the base model on multiple key metrics, demonstrating its excellent performance in practical business applications.

The core technical framework of this system includes dual-driven data and knowledge, self-optimizing training, a four-stage training pipeline, and a multi-Agent collaboration mechanism. These innovative technologies enable WOWService to maintain high levels of service quality and compliance in complex and changing business environments. The multi-Agent collaboration mechanism enhances the adaptability and user experience of the overall system through cooperation between the main agent and multiple specialized sub-agents, ensuring real-time response and execution of critical information.

Technical Report: https://arxiv.org/pdf/2510.13291

Key Points:

🌟 The WOWService system significantly enhances the capabilities and professionalism of intelligent customer service through technologies such as multi-agent collaboration and reinforcement learning.  

💡 In practical applications at Meituan, the system has covered dozens of business scenarios, improving user satisfaction and surpassing the base model.  

🚀 The core technical framework includes dual-driven data and knowledge, self-optimizing training, and multi-Agent collaboration mechanisms, ensuring high-quality service delivery.