I want to roll out Google Cloud Document AI so I can pull clean, structured data from every invoice that lands in our inbox. The focus is on extraction only—no manual keying—so accuracy and reliability matter more than speed of delivery. Scope of work • Set up and configure the Invoice Parser (or build a custom processor if you think it will outperform the out-of-the-box template). • Capture all critical fields: invoice number, invoice date, vendor name and address, individual line-item details (description, quantity, unit price), subtotals, tax, and grand total. • Build an integration layer—Python, Node, or Java is fine—that uploads PDFs or images to Document AI, receives the JSON response, maps the fields to our existing ERP via REST API, and logs confidence scores. • Implement validation: flag low-confidence fields, offer a simple review queue, and retry failed calls automatically. • Deliver deployment assets (Terraform scripts or gcloud commands), source code, and concise setup documentation so my internal team can maintain and extend the pipeline. Acceptance criteria – At least 98 % field-level accuracy on a mixed test set of 100 vendor formats – End-to-end processing time under 10 s per document – Secure handling of all data (service-account IAM, no hard-coded secrets) – Clean, commented code that passes linting and unit tests If you have prior experience with Document AI (especially the Invoice Parser) and can suggest ways to fine-tune models or optimise cost, let’s talk—this will likely evolve into ongoing work once the initial integration is stable.