Researchers use LLMs to automate building code compliance checks from floor plans

Researchers have introduced a new AI-based framework to automate the tedious process of checking residential floor plans against building codes. Published as a preprint on arXiv, the work by Subash Gautam, Debaditya Acharya, Alexandra Kleeman, and Sarah Foster targets Australian design policies such as SEPP65, BADS, and SPP7.3, which mandate specific geometric and spatial requirements for daylight, ventilation, privacy, and space efficiency. Currently, compliance checks are done manually, making them time-intensive and difficult to scale across thousands of apartment units.

The proposed framework combines three core engines. First, a Rule Engine leverages a Large Language Model (LLM) to convert text-based building regulations into executable, explainable rules. Second, a Data Extraction Engine uses computer vision to segment floor plan images into elements (walls, rooms, fixtures, text, symbols) and transforms them into a structured building graph preserving topological relationships. Finally, a Compliance Check Engine evaluates this graph against the LLM-derived rules. This approach promises a scalable, consistent, and transparent alternative to manual checks, helping enforce healthier urban development standards across jurisdictions.