Speaker
Description
The rapid growth of space science missions has led to an unprecedented increase in scientific data, engineering documents, metadata, and software resources. Efficiently organizing these heterogeneous resources and enabling intelligent access have become critical challenges for modern scientific research. At the National Space Science Data Center (NSSDC)， Chinese Academy of Science, we are developing an AI-agent framework to support intelligent scientific data management, retrieval, and analysis for space science applications.
Our framework adopts a multi-agent architecture in which specialized agents collaborate to accomplish complex scientific tasks through planning, reasoning, and tool invocation. Existing scientific data processing and analysis software have been encapsulated as Model Context Protocol (MCP) services, enabling LLM agents to directly discover and invoke domain-specific tools without requiring customized integrations. This service-oriented architecture significantly improves the flexibility and scalability of AI-assisted scientific workflows while allowing legacy scientific software to be seamlessly incorporated into agent-based systems.
To enhance knowledge acquisition, we have also developed a Retrieval-Augmented Generation (RAG) framework that integrates heterogeneous scientific resources, including mission documents, metadata, technical reports, scientific publications, and archived datasets. Beyond conventional text retrieval, the system supports multimodal knowledge retrieval from both textual descriptions and scientific images, providing more accurate and context-aware responses to researchers. The combination of semantic retrieval and domain knowledge substantially improves the efficiency of scientific information discovery and reduces the effort required to locate relevant datasets and documentation.
The integration of multi-agent collaboration, MCP-enabled scientific software, and multimodal RAG establishes a unified intelligent research environment for space science.