Data for document AI and OCR
Documents that resist parsing, captured and structured so extraction models can learn from them.
What Document AI / OCR needs
Document models struggle on creased receipts, handwritten forms and unusual layouts, which are exactly what production sends them.
We collect real documents across scripts and layouts, then annotate words, fields, tables and relationships, with redaction verified where documents contain personal data.
Data types
- Image
- Documents
- Structured output
Industries
- Banking and finance
- Retail and e-commerce
- Healthcare
- Real estate
- Telecom
- Logistics
The data it takes
Printed and handwritten text
Images of real documents in the scripts you need.
Word and line boxes
Text regions marked for OCR training.
Key-value pairs
Fields such as totals, dates and names linked to their values.
Table structure
Rows, columns and cell contents captured faithfully.
Redacted copies
Documents with personal data removed and verified.
How we collect it
- 01
Real documents, photographed
Contributors capture receipts, invoices, forms and signboards with their own phones.
- 02
Layout variety planned
A coverage plan across templates, scripts, lighting and quality.
- 03
Annotated in a layout-aware tool
Pre-labels from layout models, corrected by annotators.
- 04
Redaction verified
A second pass confirms that personal data is removed.
Related datasets
Retail product recognition
Supermarket shelves and individual products with tight bounding boxes.
Learn more
Receipt and price tag OCR
Crumpled receipts, shelf labels and price tags with word boxes and key-value links.
Learn more
Financial document OCR
Invoices, statements and forms with word boxes, key-value links and table structure.
Learn moreTell us about your Document AI / OCR project.
Describe the model and what it gets wrong today. You get volumes, languages, a quality target and a price, usually inside two working days.
- Reply within one working day
- A pilot batch you accept before we scale
Questions we get asked
Which scripts do you cover?
Latin and a wide range of Indian scripts, plus others on request. We confirm coverage and timeline during scoping.
What output formats are available?
COCO, PubTabNet-style tables, JSON key-value output and your own schema.
