An LLM-Based Framework for Automated Recognition and Structuring of Construction Cost Estimates

The construction industry continues to encounter considerable inefficiencies associated with the manual interpretation and processing of heterogeneous cost estimation documentation, which is characterised by a paucity of background data. The present study puts forward a Large Language Model (LLM)-based framework for the automated recognition, semantic structuring, and enrichment of construction cost estimates originating from unstandardised Excel spreadsheets. The proposed framework uses a combination of domain-specific data annotation, machine learning-assisted document understanding and semantic similarity retrieval to convert semi-structured procurement and estimation documents into standardised, machine-readable datasets. A large corpus of historical construction estimates was collected, organised and manually labelled to create a training dataset for the LLM-based recognition pipeline. An API-oriented architecture automatically parses, classifies and converts uploaded spreadsheets into structured entities, including estimate positions, quantities, units, unit costs and total costs. The framework also incorporates a semantic similarity matching mechanism that compares newly uploaded estimates with historical priced estimates stored in the reference database. Consequently, the system can automatically suggest historically relevant unit prices and related CO₂ indicators for similar construction positions. Experimental validation has demonstrated that the proposed framework achieves a semantic position-matching accuracy of over 85%, enabling the automatic completion of 50–70% of newly uploaded estimates. Furthermore, the solution reduces the effort required for manual estimation by around 70% compared to traditional workflows. The findings suggest that using LLM-assisted recognition and historical knowledge retrieval can substantially improve the productivity, consistency and sustainability of construction cost management and procurement processes.