AI-Powered Automated Invoice Processing System with End-to-End Email Integration, OCR Extraction, and Real-Time Dashboard Monitoring

Abstract

This paper presents the design, implementation, and experimental evaluation of a fully integrated AI-powered Automated Invoice Processing System that extends prior workflow-stage research into a complete, production-ready solution. Building upon our Phase 1 prototype — which demonstrated the viability of using Large Language Models (LLMs) and Optical Character Recognition (OCR) for structured invoice understanding — this final system introduces end-to-end email integration via Microsoft Outlook, intelligent attachment classification, SQL-based centralized data storage, and a real-time React.js monitoring dashboard. The system autonomously fetches invoice emails, classifies attachments (PDF, image, plain text), extracts key invoice fields using a hybrid OCR-LLM pipeline, validates extracted data through a rule-based correction engine, and logs all records into a structured SQL tracker. Experimental results across 510 diverse invoice samples demonstrate a field extraction accuracy of 97.4%, total computation accuracy of 99.1%, and an average processing time of 1.8 seconds per invoice. The system achieves superior performance compared to traditional OCR tools, Layout LM-based models, and our own Phase 1 AI Invoice Builder baseline, while also supporting multilingual processing, irregular layout handling, and privacy-preserving hybrid cloud deployment.