Integrating Large Language Models with RPA: A Governed Neuro-Symbolic Architecture for High-Compliance Automation

This paper proposes a governed AI–RPA architecture that formally separates probabilistic reasoning from deterministic execution for high‑compliance document processing. While large language models improve interpretation of unstructured data, their non‑deterministic nature raises audit and governance concerns in regulated industries. The proposed architecture positions UiPath as the orchestration and control plane, with Gemini acting as a bounded reasoning service. A contractor insurance compliance case study demonstrates significant reductions in manual review effort while preserving auditability and regulatory controls.