Introducing Generative Key-Value Extraction in OCI Document Understanding

OCI Document Understanding now includes a new custom model type powered by generative extraction using Large Multimodal Models (LMMs). This feature streamlines large-scale document processing for use cases such as invoices, purchase orders, résumés, fraud detection, forms, and more. You can define extraction fields through natural language prompts, leading to improved accuracy without extensive training.

Key Features

• Extracts structured data into normalized JSON format using multimodal vision models.
• Handles multi-page documents, mixed layouts, multilingual content, semi-structured, and unstructured formats.
• Can enhance accuracy by incorporating a few examples.
• Applies built-in preprocessing and post-processing logic to help with stable outputs and minimized hallucinations.
• Integrates with existing Document Understanding Custom Key-Value (KV) workflows, requiring no changes to pipelines.

Availability

This feature is available in the following regions: Chicago, London, Osaka, and São Paulo.

For guidance, see the Document Understanding documentation.