Abstract: To enable railway marshalling yard staff to quickly query production data and regulations, this paper designed an intelligent inquiry system for marshalling yard based on the domestic large language model DeepSeek-R1. The paper constructed a natural language interactive service by localizing the private deployment of DeepSeeker R1 and securely integrating it with the Computer Integrated Process System (CIPS). By combining layered agent architecture, hybrid retrieval, and Natural Language to Structured Query Language (NL2SQL) technology, the designed system supported staff to obtain real-time production data and regulatory knowledge such as train status and shunting plans. The application shows that this system can accurately answer the questions raised by users and provide technical support for the intelligence of railway freight transportation.