Abstract: In the digital office scenario of railways, some procedural and repetitive basic transactions occupy a large amount of human resources, make it difficult for teams to fully unleash their potential in specialized and high-value fields. Therefore, this paper carried out the architecture design and key technology research of a railway intelligent office system based on large models. It analyzed the current status of railway digital office systems and the problems in current office scenarios, focused on key technologies such as speech transcription that integrates multi person voiceprint recognition, railway knowledge question answering based on retrieval enhancement, multimodal data analysis based on graphic and text understanding, and text generation based on thinking chains. It also provided practical solutions for the application of large models in the field of railway intelligent office. The practical results have shown that the research achievements have effectively improved the efficiency and management level of railway office, enhanced the intelligence level of the office field, provided new impetus for the development of new quality productive forces in railways, and also provide reference for further application research of intelligent technology in the railway office field.