Announcing OCI Document Understanding custom model support

June 14, 2023 | 4 minute read
Kate D'Orazio
Principal Product Manager
Text Size 100%:

The Oracle Cloud Infrastructure (OCI) Document Understanding service allows you to perform sophisticated document extraction at scale without any data science background. It provides pretrained models for text extraction with optical character recognition, document classification, table extraction, and key value extraction for invoices, receipts, passports, and driver licenses.

Today, we’re excited to announce the general availability of two new features: Custom key value extraction and custom document classification. With these additions, customers can easily tailor key value extraction and classification models to their industry or use-case specific documents. These features are available in all OCI commercial regions and can be accessed through OCI software developer kits (SDKs) and REST APIs.

Custom key value extraction for Document Understanding

Imagine that you’re in charge of processing bill of lading forms for your logistics company. Each day, you receive hundreds of these forms containing crucial information, such as tracking IDs, recipient and carrier names, and truck assignments. Traditionally, your team has been manually extracting this information from the forms and entering it into your system, which is time-consuming and prone to errors. This tedious work affects employee morale and productivity, and with the increasing volume of forms, it’s becoming a growing challenge to manage.

Using OCI Document Understanding, you can create a custom key value extraction model that learns from the work done by humans in the past. When the model is trained on previous bill of lading forms, it can identify these key fields on incoming bill of lading documents automatically, freeing humans to perform less tedious tasks.

A screenshot of the Analyze page in the Console showing the model details for a custom key value extraction model.

Custom document classification for Document Understanding

With custom key value extraction, we’re also introducing document classification as a new feature. This powerful tool enables you to automatically categorize documents based on their layout and content, making it easier than ever to manage and process large volumes of documents. Imagine that you’re hiring new employees and require applicants to submit proof of identity documents, vaccination records, recommendation letters, and a completed application form. Using custom document classification, you can automatically categorize incoming documents into these categories to automate tasks like asking the applicant for a missing document, validating names are consistent among IDs and certificates, forwarding a recommendation letter to the hiring manager, and many other tasks.

A screenshot of the Analyze page in the Console showing the model details for a custom document classification model.

Using custom document understanding models

To train a custom model using OCI Document Understanding, use the following steps:

  1. Start with a labeled dataset. You can easily label documents with the OCI Data Labeling service and follow tutorials videos posted in our playlist.

  2. Create a project in OCI Document Understanding. Here, you can organize a collection of models for a specific use case or task. To create a model, select the model type—key value extraction or document classification—and a training duration. The default is “recommended.”

  3. When a new model training has begun, model training progress, logs, and final quality metrics are available on the Model Details page in the Oracle Cloud Console and through the Document Understanding model API.

  4. To call your custom document model, include the model OCID as part of the modelID field in your input request. The following example shows a JSON request to call a custom key value extraction model:

    "processorConfig": {
    "processorType": "GENERAL",
    "features": [
    "modelId": "[your-model-ocid]",
    "featureType": "KEY_VALUE_EXTRACTION"
    "isZipOutputEnabled": false
    "compartmentId": "[your-compartment-ocid]",
    "inputLocation": {
    "sourceType": "INLINE_DOCUMENT_CONTENT",
    "data": "......"
    "outputLocation": {
    "namespaceName": "[your-namespace]",
    "bucketName": "[your-bucket]",
    "prefix": "[your-prefix]"

Real document understanding use cases

Thousands of foreign trade documents, such as commercial invoices, bills of lading, and origin certificates, are processed by a single enterprise every month. Manual receipt and processing of such documents lead to failure, low compliance, and potential data errors. Dealing with large amounts of documents and scale business’ operations globally is a challenge.

eCOMEX NSI built a document extraction and automation solution using OCI Document Understanding called eCOMEX Digital Assistant. By using custom key value extraction, they were able to extract supplier and recipient information, pricing, and merchandise data fields from foreign trade documents. As a result, they were able to minimize significantly or eliminate manual inputs of data. By automating customer’s routines in handling foreign trade documents, they were able to reduce costs, streamline operations and assure compliance.

The OAN Order Management application provides a workspace where people, processes, and technology come together. Cutting-edge Oracle Cloud technology is used to greatly simplify and automate mundane tasks so that the Customer Service team can better focus their time on value-added activities that increase their customer engagement. oAppsNET used Oracle AI Document Understanding because it’s easy to learn, build a custom AI model, and integrate into their sales order automation solution, which allowed oAppsNET offer an enterprise sales order automation solution with minimum time and effort.

Get started today

We encourage you to try these exciting new capabilities. They’re available to you from the Console and through various SDKs and REST APIs. For more information, see the following resources:

  • Try the custom model tutorial

  • Documentation

  • Contact us at

Kate D'Orazio

Principal Product Manager

Previous Post

Transforming the organ matching and transplant journey with OCI Data Platform

Rajib Ghosh | 13 min read

Next Post

Deploying Large Language Models with OCI Data Science Using NVIDIA GPUs and Triton Inference Server

Tzvi Keisar | 4 min read