Harnessing Large Language Models for Digital Building Logbook Implementation

Digital Building Logbooks (DBLs) have been proposed to preserve lifecycle data across the design, construction, operation, and renovation phases of buildings. Yet, implementation has been hindered by the absence of standardized data models across jurisdictions and stakeholder practices. This paper argues that Large Language Models (LLMs) offer a solution that reduces reliance on rigid standardization. To test this approach, we first draw on parallels from the healthcare sector, where LLMs have extracted structured information from unstructured electronic health records. Second, we present an LLM-based workflow for processing unstructured building inspection reports. The workflow encompassed three tasks: (1) qualitative summary, (2) quantitative summary, and (3) risk level assessment. Sixteen inspection reports were processed through GPT-4o across 320 runs via a Python script. Results showed perfect consistency for categorical fields and Boolean indicators, minimal variability for ordinal severity ratings (σ ≤ 0.6), and stable risk assessments with 87.5% of reports showing low standard deviations. Each report was processed in under 10 s, representing up to a 100-fold speed improvement over manual review. These findings demonstrate the feasibility of post hoc standardization, positioning DBLs to evolve into large-scale knowledge bases that can substantially advance research on the built environment.