Hybrid RAG Approach for Building Permit Applications

This technical report presents a hybrid Retrieval Augmented Generation (RAG) approach aimed at generating ontology-based information containers from building permit application submissions. The study outlines an end-to-end method that combines Large Language Models (LLMs) with a hybrid RAG layer to extract relevant information from heterogeneous and unstructured data typically found in building permit applications. The proposed pipeline is designed to produce standard-compliant information containers in accordance with ISO 21597, facilitating the classification and linking of documents to application processes. The report details the challenges faced, including scalability and the handling of ambiguous regulatory language, while also discussing the potential of LLMs in automating and enhancing semantic integration within digital building permit processes. The methodology includes a proof-of-concept implementation that demonstrates the effectiveness of the approach in populating OntoBPR entities and generating metadata for formal checks of application completeness.