Enterprise · AINeural Software Solutions, SRL

OCR Invoice Ingestion Pipeline

AI-powered purchase invoice registration for the accounting team. Upload a photo or PDF through a custom web form, get a fully created Purchase Invoice in ERPNext — with no file ever touching the server. Azure Document Intelligence handles extraction; a Logic App routes the result, auto-creating suppliers from the DGII registry when they don't already exist.

Architecture

System flow

BROWSERAZUREDGIIERPNEXTWeb form — invoice uploadDrag-and-drop · validated in browser · no server storageAzure Logic App — invoice pipelineReceives image · orchestrates extraction and registrationAzure Document IntelligenceAsync image analysis · polling loop · up to 3 minInvoice field extractionRNC · NCF · date · totals · DGII verification URLfoundnot foundSupplier foundMatched by RNC in ERPNextDGII registry lookupQuery by supplier RNCSupplier auto-createdRegistered in ERPNextPurchase Invoice createdResult returned to user · success link or error message in SpanishStructured error codes: duplicate invoice · missing RNC · unrecognized supplierBrowser / ERPNextAzureDGII
Interface

In action

The custom web form handles five distinct states — each with clear Spanish-language feedback tailored to the accounting team.

Empty state
Empty state
Drag-and-drop upload · client-side file validation
Processing
Processing
AI extraction in progress · 1–2 min estimate shown
Success
Success
Invoice created · clickable link to ERPNext document
Error
Error
Extracted data shown · actionable recovery tips in Spanish
Duplicate detected
Duplicate detected
NCF + supplier matched · links to the existing invoice
My role

End-to-end: the custom web form and client-side validation, the Azure Logic App workflows, the Document Intelligence integration, the ERPNext Purchase Invoice creation, and the Spanish-language error handling surfaced back to the user.

What it solves

Manual invoice entry is slow and error-prone. This pipeline lets the accounting team photograph a paper invoice and have it registered in seconds — including auto-fetching the supplier from the DGII registry if they've never been seen before. The model improves accuracy with every invoice processed.

Stack
Azure Document IntelligenceAzure Logic AppsAngularERPNextPython