Abstract: The advancement of Industry 5.0 imposes higher demands on the informatization, digitization, and intelligence of manufacturing. Addressing challenges such as the heavy reliance on expert experience in traditional industrial application development and the hallucination traps prevalent in existing general-purpose large model frameworks, a rapid construction method for industrial applications leveraging domain-specific large language models (LLMs) and collaborative agents is proposed. The methodology begins by constructing a knowledge graph for industrial application development, which is then embedded into a prompting framework to establish a domain-specific LLM tailored for rapid industrial development. Subsequently, the LLM is decomposed into multiple collaborative agents, including requirement analysis agents, coding agents, and testing agents, enabling an integrated workflow for accelerated industrial application development. To validate the effectiveness and superiority of the proposed method, a subjective and objective comprehensive evaluation was conducted using real-world development cases, comparing the proposed approach with traditional methods. Results demonstrate that the proposed method not only exhibits enhanced development capabilities but also provides more user-friendly human-machine interaction, making it better suited for contemporary rapid industrial application construction services.

Key words: domain-specific large language models, multi-agents, industrial applications, knowledge graph