AI & Automation / Case study
An open-source workspace that turns invoice documents into structured, reviewable data, combining Azure Document Intelligence extraction with human review.
Open-source project. Screenshots show the demo interface with fictional documents and extraction results.
01 / The challenge
Turning PDF and image invoices into structured records requires more than extracting text. Uncertain fields need to be checked against the original, corrections must survive a new analysis, and users need a clear view of what is processing, ready for review, or saved.
02 / The approach
Built a Next.js and TypeScript workspace backed by an ASP.NET Core API and PostgreSQL. Azure Document Intelligence extracts invoice headers and line items, while RabbitMQ and a .NET background worker handle analysis asynchronously. The review screen pairs the original with editable fields and confidence indicators. Draft recovery preserves edits during reanalysis, and Google sign-in with API ownership checks isolates each user's documents.
03 / The outcome
Implemented the workflow from document upload through extraction, human review, and invoice saving, with CSV export of saved invoice headers. An account-free demo uses fictional examples to make the review experience explorable. Automated tests cover workflow and extraction regressions; demo screenshots illustrate the interface rather than proving live service integration.
