From Ambiguity to Execution: An Agentic Neuro-Symbolic Framework for Transforming Building Regulations into Deterministic Constraints

Abstract

Integrating Large Language Models (LLMs) into Automated Compliance Checking (ACC) introduces "Spatial Hallucinations" and "Serialization Bottlenecks" when processing massive Building Information Models (BIM). This research proposes an Agentic Neuro-Symbolic Framework that decouples semantic interpretation from geometric verification. Instead of relying on generative LLMs for spatial reasoning, an agentic orchestrator synthesizes dynamic logic for a deterministic geometry kernel (IfcOpenShell). Evaluated against the Australian National Construction Code (NCC 2022), the framework autonomously resolves ISO 16739 ontological ambiguity. By employing connectivity graph traversal, it isolates targeted structural sub-graphs, reducing computational complexity from O(N) to O(K) and bypassing context-window limits. Offloading spatial mathematics to this deterministic environment achieves a zero-hallucination rate, generating immutable audit trails for digital permitting. This establishes a highly scalable, context-dependent foundation, proving AI in engineering must orchestrate deterministic tools rather than probabilistically predict physical realities.