AI-Powered Invoice Processing at Scale on Azure

Designed and built a scalable Azure-based system for processing large volumes of PDF invoices and converting them into structured XML.

The solution uses an event-driven architecture with Azure Blob Storage, Event Grid, Queue Storage, Azure Functions, AI Document Intelligence, SQL Database, and Application Insights.

A key design decision was to avoid sending every invoice through AI. Known invoice formats can be processed with deterministic Python parsers, while Azure AI Document Intelligence handles unknown or complex documents. This hybrid approach can significantly reduce AI API usage while keeping the system flexible.

I also built a lightweight dashboard for uploading invoice batches, tracking processing status, and reviewing results independently from the processing pipeline.

The project focuses on scalability, reliability, observability, and cost-efficient use of AI.