Invoices and receipts
Arabic-English line items, totals, taxes, discounts, and stamps, with the scan quality and layout variation your system encounters.
Bayanat Documents
Arabic documents with the labels your model needs.
Bayanat Labs (Riyadh) builds Arabic document datasets to your scope, from bilingual invoices to handwritten forms, with text, layout, fields and evidence labeled by annotators from a team of 50,000+ expert annotators.
اقرأ بالعربيةWhich Arabic document data fits your Saudi or UAE project?
Start from what your model has to do. Bayanat Documents is built to order, and the open releases linked here let you test a pipeline first; they are not samples of custom delivery. Saudi and UAE invoice rows come first, and the rest apply in any market.
| Your project | Start with | Check before you commit |
|---|---|---|
| Extracting fields from Saudi invoices and receipts | Custom invoice scope and acceptance Synthetic Saudi E-Invoices (open, for pipeline tests) | Whether the scope covers the layouts and scan quality you actually receive, whether test issuers and templates stay out of training, and whether each value is tied to its place on the page. The open set is fully synthetic, for research, and its pages are not valid tax invoices. |
| Building a gold set of UAE invoices and Arabic-English tables | Custom gold set scope | List the fields you need, such as tax registration numbers and line items, plus table and layout labels. Bayanat publishes no UAE invoice set, so sources come from documents you can authorize. Confirm the hosting location in the scope: in-region is the default, with private-cloud and on-premises options. |
| Reading Arabic forms, handwriting and checkboxes | Custom form labels Arabic Form Field Extraction (open, synthetic) | How handwriting, corrections and missing entries are labeled in your own forms. The open set is synthetic with controlled language and has no handwriting or scan degradation, so it cannot show accuracy on real forms. |
| Licensing a ready-made library of genuine Arabic invoices | Scope a custom collection instead | This family has no stocked library. From any supplier of stock documents, ask for issuer and template counts, source permissions, and which pages are genuine, de-identified or synthetic. |
| OCR text and reading order only | Arabic OCR annotation services 10-point Arabic OCR dataset audit | Render right-to-left lines beside their crops on a sample you draw yourself, and check mixed Arabic and Latin lines and digits before accepting a batch. |
| Answering questions from contracts and procurement files | Custom contract and procurement scope Bayanat Search (finding passages across a corpus) | Whether each answer points to a specific supporting passage, how clauses, tables and attachments are labeled, and which documents you are authorized to share. Add Search if the system must first find the right passage across a document set. |
| Testing a model on documents it has not seen | Bayanat Bench Custom document test splits | Held-out documents kept separate from training deliveries, with access and versioning defined for the project. In a custom document set, evaluation splits can separate issuers, templates, sources and writers, with template diversity reported separately from page count. Acceptance thresholds stay your decision. |
| OCR research on a large public set of Arabic pages | Arabic Document Atlas (200,286 images, open) | Most pages come from older books, not business documents. The OCR is machine-made with no measured error rate, word positions are approximate, and rights differ by component. |
Is Bayanat Documents the right fit?
A good fit if
- Your model reads Arabic or bilingual invoices, receipts, forms or contracts.
- You need more than OCR text: reading order, layout, extracted fields and the evidence behind them.
- You need test splits by issuer, template or writer, so you can measure performance on unfamiliar documents.
Choose something else if
- You want a ready-made document image library. This family is built to order.
- You need OCR software. We build the training and evaluation data, not the OCR tool.
What you can check before you commission
We separate what is confirmed on this page, what we send on request, and what is agreed for each project.
| Item | Status | Details |
|---|---|---|
| Volume | Per project | Built to order. Page count and template coverage are agreed with you in a fixed-scope pilot with gold-standard QA and a quality report, usually within two weeks. |
| Label types | Per project | Transcription, coordinates, reading order, layout, normalized fields, evidence locations and validation checks; choose what your model needs. |
| Source handling | Confirmed | Genuine, de-identified and synthetic documents remain distinguishable. Template diversity is reported separately from page count. |
| Evaluation splits | Per project | Separate issuers, templates, sources and writers for testing. |
| Example of the label format | On request | Ask for it when you scope your pilot. |
What we can build together
Forms and records
Printed fields, handwriting, checkboxes, corrections, and missing entries in applications, delivery documents, or service reports.
Contracts and procurement
Clauses, tables, attachments, and cross-references, with questions answered from specific supporting passages.
Labels beyond transcription
Choose reading order, layout, normalized fields, evidence locations, and checks for missing information or inconsistent totals.
How the license covers each use
Training rights do not automatically include display, retrieval or redistribution. Each use is set in the agreement.
Source documents
Source permissions are defined with you before collection.
Delivered data
Use and delivery rights are set in the agreement.
How we put the data together
Start with one document type
We define source permissions, layout diversity, capture conditions, and page count around the problem you need to solve.
Keep sources identifiable
Genuine, de-identified, and synthetic documents remain distinguishable. Template diversity is reported separately from page count.
Test unfamiliar documents
Evaluation splits can separate issuers, templates, sources, and writers to check performance on documents the model has not seen.
For example: Checking an invoice total
A bilingual invoice has a discount and a handwritten correction. Labels locate both amounts so the system can extract them, recompute the total, and flag a mismatch.
Scope your request
Tell us these details. We come back with a fixed-scope pilot plan on your own data: gold-standard QA with adjudication and rework loops and a quality report, usually within two weeks, hosted in-region by default.
- Document types
- Languages and scripts
- Labels you need
- Sources you can authorize
- Volume and timeline
Common questions
Can I order OCR labels without extraction labels?
Yes. You can request transcription, coordinates, and reading order on their own, or add layout, field extraction, document questions, and validation checks. The labels should match the model you are building.
How do you handle repeated invoice or form layouts?
For a custom collection, we agree the template coverage and evaluation splits with you. Variations of one template remain identifiable, so page count does not get mistaken for the number of independent layouts.
Do you sell a ready-made Arabic invoice dataset?
Bayanat Documents is built to order, so there is no stocked invoice library to license. To test a pipeline first, the open Synthetic Saudi E-Invoices, Receipts and Bilingual Forms release has 25,000 generated records under CC BY 4.0; it is synthetic, for research, and its pages are not valid tax invoices. For real layouts, we scope a custom set from documents you can authorize.
Can you build a gold set of Saudi or UAE invoices?
Yes, as a custom project. We agree document types, fields, table and layout labels, source permissions and evaluation splits in a fixed-scope pilot with gold-standard QA and a quality report, usually within two weeks. Genuine, de-identified and synthetic documents stay distinguishable. Confirm the hosting location in the scope: in-region is the default, with private-cloud and on-premises options.