🤖 AI Summary
This work addresses the challenges posed by the heterogeneous multimodal nature of enterprise policy documents, which often cause large language models to hallucinate, disrupt table structures, and lack end-to-end controllability—resulting in labor-intensive manual processing requiring 2–3 days per document. To overcome these limitations, the authors propose a governed multi-agent collaboration framework grounded in a shared, versioned rule repository. The framework integrates large language models (LLMs), vision-language models (VLMs), schema validation, and human-in-the-loop mechanisms through six specialized agents that collaboratively perform parsing, multimodal extraction, consistency verification, evaluation, iterative refinement, and personalized artifact generation, while ensuring full traceability across the pipeline. Evaluated on 120 real-world documents, the approach achieves a 96% success rate, automatically extracts 3,896 rules (71.4% auto-approved), produces 812 deployable artifacts, and reduces per-document processing time to 40–125 minutes.
📝 Abstract
Enterprise guideline documents are heterogeneous and multimodal, combining narrative text, complex tables, and embedded images. Existing LLM and VLM systems face hallucinated content, table structure degradation, and lack governed workflows extending beyond extraction to validation and artifact generation. This leaves enterprises to perform this manually, consuming 2-3 days per document. To address this, we introduce GUIDE, a governed multi-agent framework built on a shared versioned rule store with schema-validated inter-agent contracts and end-to-end provenance tracking. Six specialized agents handle parsing, VLM-driven extraction, consistency checking, evaluation, human-in-the-loop (HITL) escalation, and persona-tailored artifact synthesis. Evaluated on 120 real-world enterprise guideline documents, GUIDE achieves 96% document success, extracts 3,896 rules with 71.4% auto-approved, produces 812 deployment-ready artifacts, and reduces turnaround to 40-125 minutes per document.