Chat with Pre-Demolition Audits: Knowledge-Graph Based Information Retrieval

Abstract: Although pre-demolition audits (PDAs) provide early information on building materials, they are underutilized partly because their data require domain expertise to access and interpret. This raises the question: can natural-language querying enable non-experts to retrieve information from PDA datasets? We propose a knowledge-graph-based Retrieval-Augmented Generation (RAG) framework for natural-language querying of PDAs, applied to a case study of 40 French audits. Evaluated against a semantic RAG baseline across 21 queries, Graph-RAG achieves a success rate of 90% compared to 57% using vector-based retrieval, though at 2.34× higher cost, demonstrating its potential to making PDA content accessible to non-expert stakeholders.