User manual
How to test invoice recognition with this demo. Nothing is sent to Azure until you press Process.
1. Account
Anyone can register with an email address and a password (at least 8 characters). Log in afterwards with the same details. Everything you create is private to your account.
2. Projects
A project stands for one company whose invoices you want to process, for example Acme Trading. Create it on the Projects page. A project holds any number of batches. Deleting a project deletes all of its batches and files.
3. Batches and the waiting area
Inside a project, create a batch and give it a name such as March Invoices. A new batch is a draft: its files sit in a waiting area on this server and are not processed yet.
- Drop files on the upload box or choose them, then press Add to waiting area.
- Maximum 10 files per batch, 20 MB each. Accepted types: bmp, jpeg, jpg, pdf, png, tif, tiff.
- Each file is shown with its page count and size. Remove a file with the ✕ button while the batch is still a draft.
4. Processing
Press ▶ Process. The batch turns to processing and the documents are sent to Azure AI Document Intelligence (prebuilt-invoice model) one after the other. The page updates by itself: a progress bar, the file currently being read, and per file the status, vendor, invoice number, total and the time it took.
- A file that Azure rejects is marked failed with the reason. The others continue.
- When every file is finished the batch is done. Use Retry failed documents to send only the failed ones again, or Process again (all) to redo the whole batch.
- You can leave the page during processing. It carries on in the background and the batch page shows the current state when you come back.
5. Viewing the results
Click a file name in a finished batch. The viewer shows the page image on the left and the fields that were read on the right, grouped into Vendor, Customer, Invoice, Addresses, Totals, Line items and Tax details.
- Every value carries a confidence badge: ≥ 80% 50–79% < 50%.
- Hover a field to highlight where on the page it was read; hover a box on the page to highlight its field.
- Previous / Next step through the batch. Original downloads the uploaded file and Raw JSON opens the complete Azure response.
- The full text recognised on the page is available at the bottom of the viewer.
6. Fields that are read
Only fields the model actually finds on the invoice are shown.
| Vendor | VendorName, VendorAddress, VendorAddressRecipient, VendorTaxId |
|---|---|
| Customer | CustomerName, CustomerId, CustomerAddress, CustomerAddressRecipient, CustomerTaxId |
| Invoice | InvoiceId, InvoiceDate, DueDate, PurchaseOrder, PaymentTerm, ServiceStartDate, ServiceEndDate |
| Addresses | Billing, Shipping, Service and Remittance address plus their recipients |
| Totals | SubTotal, TotalTax, InvoiceTotal, PreviousUnpaidBalance, AmountDue, TotalDiscount |
| Line items | Description, Quantity, Unit, UnitPrice, ProductCode, Date, Tax, Amount per line |
| Tax details | Rate and Amount per tax line |
7. Good to know
- Uploaded files, page previews and results are stored on this server and can be removed by deleting the batch or project.
- Files are only transmitted to Azure when you press Process, and only the batch you pressed it for.
- If the server is restarted while a batch is processing, the unfinished files go back to the waiting area and can be processed again.
- Best results come from clear scans or PDFs of complete invoices. Multi-page PDFs are supported; previews are made for the first 10 pages.