Rossum has two types of AI Engine: Generic AI Engine and Dedicated AI Engine.

Rossum has several types of Generic Engines that have learned to process a specific document type. These include:

• Accounting Engine (Accounts Payable and Receivable Engine)
• Chinese Invoice Generic Engine
• Certificate of Analysis Generic Engine

Generic Engines learn to recognize fields from many various invoice layouts, languages, and types of content. Generic Engines are default for standard Rossum accounts.

Why Generic AI Engine?

Rossum’s Generic AI Engine is suitable if you:

• Process documentation with standard layouts
• Need to extract data that Rossum can extract by default
• Are satisfied with the accuracy and confidence levels.

For simple usage, generic AI engines can achieve moderate to high levels of document data capture accuracy. If your process complex and highly diverse documents, you may need a Dedicated AI Engine.

Why Dedicated AI Engine?

Dedicated AI Engine learns from your unique, business-specific data. As you review, annotate, and validate documents via the Rossum user interface, you are automatically providing the engine with valuable data for learning. Your annotations increase data extraction accuracy for your documents.

Rossum’s Dedicated AI Engine is suitable if you:

• Process documents with custom, diverse layouts.
• Receive many documents in different languages.
• Need to extract custom Data fields that the generic AI engines do not extract by default.
• Want to continuously increase levels of data capture accuracy.

What Data is Used for Dedicated Engine Training?

All documents that are to be used for dedicated AI engine learning process must pass through a verification process, which is carried out through the Rossum validation screen. After Rossum processes an uploaded document, a data entry operator opens the validation screen and reviews the captured data. The operator can correct errors by pointing and clicking on the bounding box around erroneous values. Alternatively, the operator can edit values to fix typos.

Once they’ve finished their review, the operator clicks the Confirm button and can continue to the next document. This is called the annotation process.

Please keep in mind that the dedicated AI engine cannot be trained without using the Rossum validation screen. Data validation via the Rossum interface provides the engine with information that is essential to the learning process.

We have written a guide to help you better understand the annotation process, which will ensure you attain the highest levels of data capture accuracy from your Rossum dedicated AI engine.

Note: If you want to use a Dedicated AI Engine, make sure you have purchased the feature before you start the training process.