Fine-tuning generative AI for extraction

You can now fine-tune generative AI models specifically for your document types, delivering strongly improved extraction accuracy with minimal training data.

This isn't just an incremental update, it's a next step for document understanding. Here’s what it means for you:

• Up to 10x less training data: where you previously needed hundreds of examples, you can now achieve superior results with as few as 10-20 documents per type, allowing you to go live in hours, not weeks. 
• Visual understanding: the new models don't just read; they see the whole document. This visual understanding allows them to interpret information that was previously out of reach. The AI can now recognize logos to identify a supplier, understand technical drawings, or use the placement of a signature to validate a form. It comprehends the document's structure visually, just like a human would.
• Deep contextual awareness: go beyond simple key-value extraction. They understand the document context fully, allowing them to make logical inferences about data relationships in e.g. complex table structures while maintaining accuracy.
• Superior handwriting recognition: we've made a big leap in processing handwritten text. The new models are significantly more adept at deciphering even messy, cursive, or rushed handwriting. This unlocks reliable automation for a whole new class of documents, like handwritten intake forms, field service reports, and annotated logs.