AI-Powered Integrated Framework for Autonomous Invoice Generation and Intelligent Information Extraction via Distributed Multimodal Web Architectures

The global financial ecosystem is currently travers-ing a critical transition from legacy, manual-centric accounts payable workflows toward autonomous Intelligent Document Processing (IDP) systems. Traditional invoice management is characterized by significant operational friction, with manual processing costs ranging from $12 to $30 per document and error rates frequently exceeding 3%. This research paper presents the design and implementation of an AI-powered invoice generation and extraction framework integrated into a modern distributed web environment. The proposed system leverages a multimodal transformer architecture, specifically utilizing LayoutLMv3 and Document Understanding Transformers (Donut), to move be-yond simple character recognition toward deep semantic and spatial understanding of document layouts. By employing a MERN (MongoDB, Express, React, Node.js) stack coupled with a Flask-based AI microservice, the architecture facilitates the conversion of unstructured inputs—such as natural language prompts, emails, and scanned images—into structured, validated financial records. Benchmarked against the SROIE dataset, the system demonstrates an F1-score of 0.95 for key entity extraction and achieves an 80% reduction in per-invoice processing costs. The investigation further delineates a ”Human-in-the-Loop” validation mechanism and explores advanced fraud detection through anomaly detection algorithms. The results indicate that the integration of agentic AI within web-based financial tools significantly enhances operational throughput, providing a scalable solution for enterprise-grade financial automation.