RETRIEVAL-AUGMENTED CODE EXECUTION FRAMEWORK FOR AUTOMATED STRUCTURAL CALCULATION OF PRECAST CONCRETE FLOOR PANELS

Large language models (LLMs) have recently gained attention in construction engineering for their ability to interpret technical documents, extract domain-specific information, and support natural language interfaces.To mitigate hallucination and improve factual reliability, retrieval-augmented generation (RAG) techniques have been widely adopted, enabling LLMs to ground their responses in external reference documents such as design codes.However, current RAG-based applications still face challenges in performing accurate structural calculations, particularly when precise mathematical reasoning and strict code compliance are required.To address these limitations, this paper proposes a Retrieval-Augmented Generation-based framework for automated structural calculations, integrating semantic retrieval with code-based reasoning and self-verification.The proposed system consists of a vector database that stores and retrieves relevant design code provisions and structural equations, and a prompt-driven LLM that generates executable Python code along with human-readable reasoning.Furthermore, a self-verification mechanism is incorporated to ensure both logical consistency with design standards and numerical accuracy of the computed results.The framework was evaluated on precast concrete floor slab design scenarios, demonstrating that the integration of code execution (CE) and self-verification (SV) significantly improves the reliability of automated structural verification workflows.These findings highlight the potential of AI-driven approaches in supporting digital construction technologies and enhancing real-time decision-making in structural engineering.