DOCUMENT DATA FOR TRAINING, EVALS AND RL

Document data frontier AI labs can verify.

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.

CHOOSE THE OUTPUT YOUR WORKFLOW NEEDS
GritRedact

Remove sensitive information

Documents, images & audio
GritTransform

Extract structured JSON

PDFs & images
GritRender

Generate synthetic test content

Documents, images & audio

For Frontier AI Labs

Every document comes with its answer key.

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.

01

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.

02

Verifiable tasks for evals and RL

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.

03

Degraded captures and Indic scripts

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

Make success observable before production.

  1. 01Does every annotated field have a recorded value?
  2. 02Can a reward be computed without a human label?
  3. 03Which capture conditions and scripts does the model fail on?
  4. 04Are eval documents kept out of training?

Product fit

A clean division of work.

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.

01GritRedactCONTROLLED REAL DATA

Remove personal data from real documents before they reach a training set.

Detect, review, remove and verify sensitive information according to a defined downstream purpose.

  • Policy-driven
  • On-premises or SaaS
  • Auditable
Explore GritRedact
02GritTransformSTRUCTURED DOCUMENT DATA

Turn the documents you already hold into JSON.

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.

  • Custom extraction fields
  • Source verification
  • Structured JSON output
Explore GritTransform
03GritRenderSYNTHETIC EVALUATION

Generate synthetic document corpora with field-level ground truth.

Generate realistic documents, images and audio with labels, expected results and controllable variation.

  • Known ground truth
  • Edge-case coverage
  • Repeatable runs
Explore GritRender

Start with your workflow

Put your unstructured data to work.

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