Document AI

Invoice Extraction with Review Controls and Spreadsheet Handoff

A delivered invoice-processing workflow combines AI extraction, translation, validation and reviewable spreadsheet output.

The problem

Invoice headers and line items needed to move from scanned or digital documents into an existing finance workbook. Descriptions had to remain available in their original language and in English. The destination also contained calculated columns that automation must not overwrite.

What I built

  • Drive intake and n8n orchestration for individual invoices in PDF or image form.

  • AI-assisted header and line extraction, including Greek-to-English descriptions.

  • Deterministic document, VAT and total checks alongside the model output.

  • Visible success, review, duplicate and failed states.

  • Correction memory based on reviewed records, with protected identity and audit fields.

  • Google Sheets output designed around the existing Excel and Power Query handoff.

The workflow separates extraction from validation and keeps uncertain records available for human handling.

The outcome

The delivered core connects intake, extraction, review and spreadsheet output in one process. It suits operations teams that need structured invoice data while retaining control over exceptions and calculations. Maintenance and later extensions are tracked separately. No measured accuracy, throughput or savings figure is claimed; combined multi-invoice PDF processing is outside this verified scope.

Working on something similar?

Discuss a similar project →