Abstract: To address the problems of fragmented information, low efficiency of resource retrieval, and difficulties in cross-domain integration of paper documents in scientific research management, this paper designed an intelligent question-and-answer system for scientific research management based on the DeepSeek large language model. The system adopted the DeepSeek large language model as the core foundation. It completed the full structured knowledge accumulation in the field of scientific research management through knowledge extraction and knowledge graph construction, and deployed a lightweight intent understanding model to accurately identify user query requirements. The system realized full-process intelligent responses for scenarios including intelligent consultation, knowledge discovery, policy interpretation, and process navigation, with capabilities of real-time knowledge update, precise retrieval and high-quality text generation. Application results demonstrated that the system outperforms traditional retrieval tools in terms of accuracy, response speed and user satisfaction due to the integration with DeepSeek. It provides a practical technical route for the intelligent transformation of scientific research management.