Integrating large-scale language model with engineering standards and codes: Enhancing accessibility and understanding through an interactive Q&A approach

Standards and Codes in the engineering domain represent the essential benchmarks that uphold project quality to meet stipulated requirements. They embody best practices validated by extensive engineering experience and serve as pivotal assurances for project success. This paper leverages building engineering codes as the primary data source to propose a question-and-answer (Q&A) system tailored for individuals from diverse backgrounds. The system integrates information retrieval with advanced large language models (LLM), undertaking the following key tasks: 1. Given the specialized nature of engineering code documents, a specialized engineering knowledge base was constructed. Addressing the current challenge of information retrieval from figures and diagrams which contains a methodology combining ChatGPT interpretation with manual annotation by engineers was proposed. 2. In realizing information retrieval and response generation, this paper employs the RAG (Retriever-Reader-Generator) technique. Additionally, employing full-stack development with React for the frontend and Python Flask for the backend. 3. To rigorously test the effectiveness of these components, the RAGAS framework—a specialized evaluation tool for Retrieval-Augmented Generation systems—was employed. The results from this comprehensive evaluation were highly encouraging, showcasing exceptional performance in both retrieval and generation tasks. These findings not only highlight the system's capability to handle complex engineering queries but also confirm its readiness for practical deployment in real-world applications.