Training data for multimodal document models

Multimodal document models are trained on documents that mix text, tables and images. Useful records include inspection reports with photos, scanned forms, and technical drawings with notes. They show real visual variety, text tied to images, and annotated corrections.

Records for multimodal document models have personal details removed, and the supplier approves each release. A rights review runs on inspection reports, scanned forms and drawings before a sample is prepared. The sample is shared under NDA only after supplier approval.

Multimodal document models are scoped through a request describing the records, volume and uses you need. SourceX matches suitable suppliers in the U.S., Canada, the EU, the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia and runs a rights review. The license sets allowed uses, term and deletion, and resale and re-identification are barred.