AI-powered error detection in construction design review processes: an automated conceptual framework with proof-of-concept validation

Errors and inconsistencies in construction design documents cause project delays, cost overruns, and safety risks. This paper introduces a conceptual framework for an AI-powered system to detect such errors. The proposed architecture overcomes manual and rule-based limitations by synergistically combining large language models (LLMs) for semantic analysis, an advanced LLM-Optical Character Recognition (OCR) paradigm for context-aware data ingestion, and a Retrieval-Augmented Generation (RAG) framework for grounded reasoning. The primary theoretical contribution is the synergistic integration of these algorithmic components for this domain. The framework’s feasibility is demonstrated through an end-to-end proof-of-concept implementation. This case study validates the system’s ability to identify internal multimodal contradictions and external compliance violations against industry standards. Rather than relying on purely quantitative benchmarking, constrained by a lack of standardized open-source datasets, this study presents an in-depth qualitative architectural validation. The framework’s components, from data preprocessing to a Human-in-the-Loop (HITL) feedback mechanism, are detailed to provide a comprehensive roadmap for future implementation. This work bridges a critical gap between information science and construction engineering, offering a systematic approach to quality assurance.