With the continuous development of information technology, document processing has become increasingly important in the fields of law, finance, science and technology. Traditional manual processing methods are difficult to meet the requirements of efficiency and accuracy. Intelligent document workflow system based on large language model technology provides an innovative solution. Through deep learning and natural language processing (NLP) technology, it can automatically perform complex document analysis tasks and achieve efficient task flow and information extraction. This study proposes a system architecture that integrates deep learning models and workflow management engines. After experimental verification, when processing legal documents, improves the task accuracy from 75 to 92%, and improves the task completion rate from 70 to 95%. In addition, by optimizing task scheduling and model training, the system’s task completion rate is further improved to 98%. These results show that the Qwen-based system demonstrates significant advantages in processing OFD documents and complex legal workflows. Its API-centric design enables seamless integration with existing enterprise systems, while the model’s lightweight architecture ensures scalability. Future work will focus on expanding Qwen’s multilingual capabilities and enhancing real-time API performance for large-scale OFD datasets in improving document processing efficiency and accuracy, especially when processing complex documents. This study provides strong technical support for the development of intelligent document workflow systems and demonstrates its broad application prospects in industries such as electronic document exchange and mobile approval matrix.