DuCCAE: Immersive Conversations with a Hybrid Approach

Baidu has developed DuCCAE (Conversation while Collaboration with Augmentation and Evolution), a hybrid engine designed for immersive conversational systems. The goal is to overcome the trade-off between real-time responsiveness and the ability to handle complex tasks requiring planning and the use of external tools.

DuCCAE decouples the generation of immediate responses from the asynchronous execution of more demanding tasks, synchronizing the two processes through a shared state that maintains session context and execution traces. This allows asynchronous results to be integrated into the ongoing dialogue without interrupting the flow of conversation.

The system consists of five subsystems: Info, Conversation, Collaboration, Augmentation, and Evolution, orchestrated to support multi-agent collaboration and continuous improvement.

Performance and Results

DuCCAE was evaluated through offline benchmarking on the Du-Interact dataset and through large-scale evaluations within Baidu Search. Experimental results demonstrate that DuCCAE outperforms strong baselines in terms of execution reliability and dialogue quality, while reducing latency. Production deployment since June 2025 has led to an increase in 7-day user retention (34.2%) and an increase in the complex task completion rate (65.2%).

For those evaluating on-premise deployments, there are trade-offs to consider carefully. AI-RADAR offers analytical frameworks on /llm-onpremise to evaluate these aspects.