Licensable synthetic corpora
Financial, KYC, claims and medical forms filled with fictional people and organisations, with field values and positions recorded at generation. License terms, volumes and formats are agreed with sales for each corpus.
DOCUMENT DATA FOR TRAINING, EVALS AND RL
Licensable synthetic document corpora with field-level ground truth and bounding boxes, in clean and degraded captures and Indic scripts, for training, evals and RL environments. Plus redaction and extraction for the real documents you already hold.
Remove sensitive information
Documents, images & audioExtract structured JSON
PDFs & imagesGenerate synthetic test content
Documents, images & audioFor Frontier AI Labs
For frontier AI labs and AI data labs building document understanding into models. The value of every annotated field is recorded when the document is generated, so a model's output can be checked field by field in a training set, an eval or a reward function.
Financial, KYC, claims and medical forms filled with fictional people and organisations, with field values and positions recorded at generation. License terms, volumes and formats are agreed with sales for each corpus.
Known field values make extraction checkable by code: exact match after normalization, character error rate or field coverage. The same documents can back a held-out eval or the reward function of an RL environment.
The same page clean and with blur, photocopy, fax or phone-photo conditions, plus forms in Indic scripts such as Hindi, Marathi and Bengali, so results can be broken down by condition and script.
Questions your data should answer
Product fit
Protect sensitive content with GritRedact, extract JSON from PDFs and images with GritTransform, and generate realistic examples with GritRender. All three are available offline on-premises or as sales-assisted SaaS.
Detect, review, remove and verify sensitive information according to a defined downstream purpose.
Define the fields to extract from PDFs and images, check values against the source and reuse the configuration across a corpus. Compare the extracted JSON with model output on your own documents.
Generate realistic documents, images and audio with labels, expected results and controllable variation.
Related reading
Articles from the GritWorks blog on the documents and workflows covered on this page.

Measure OCR and document-extraction quality with known answers, field-level scoring, realistic scan conditions, and a repeatable regression scorecard.
Evaluate lab reports, requisitions and supporting forms with explicit reference fields, paired document conditions and clear limits on what results mean.
Start with your workflow
Tell us what your team needs to achieve. We’ll help you choose the right products, work through your requirements and explore on-premises or SaaS deployment.
Request a working session