Turning Construction Checklists into Machine-Readable Inspection Schema

Abstract: Construction verification relies on unstructured checklists, preventing integration with autonomous systems. This paper investigates how to scalable convert legacy instructions into machine-readable data. We propose a Hybrid Inspection Pipeline using LLMs and “Balanced Retrieval-Augmented Generation” to map text into an “Inspection Triad” aligned with IFC and AEC3PO ontologies. Our system achieves 92% semantic precision while eliminating hallucinations via a deterministic fidelity loop. This provides the executable logic necessary for autonomous inspection agents and establishes a foundation for future research autonomous construction inspections.