Create and publish an extraction template

Learn how to create an extraction template to extract structured data from documents in AgentFlow Unstructured.

Prerequisites:

You need an active Reltio tenant with AgentFlow Unstructured.
You must have access to AgentFlow Unstructured.
You must have a source document that contains data to extract.

Create, validate, and publish a template to define what information AgentFlow Unstructured must extract from your documents and how the extracted output maps to Reltio entities and attributes. During this process, you review the extracted JSON, verify the mapped fields, and publish the template for use in document pipelines.

To create and publish an extraction template

From Agentflow, select Agentflow Unstructured.
In AgentFlow Unstructured, select Document AI.
In the Create template section, select Let's get started. The Create new template page is displayed.
1. In the Template name field, enter a name for the template.
2. In Upload a file section, drag and drop a sample PDF file, or select click to upload a file.
3. From the Client ID dropdown, select an existing client ID or create a new client ID, and then select Fetch tenants. The Select a tenant field is displayed below. To create a new client ID, see Client Credentials at a glance.
4. From the Select a tenant dropdown, select the tenant that should receive the extracted structured data from your documents.
Select Next to proceed to the Review and Extraction step and review the extracted JSON output to confirm that the expected data was extracted.

Review the uploaded document in the left pane and the extracted JSON output on the right pane. Confirm that the extracted output contains the expected values.
1. Enter text in the Search bar to find specific fields or values in the extracted JSON output, if needed.
2. Select Finetune extraction to open the AI Chat Extraction box.
3. Add instructions for any additional data you want to extract and select Extract information. The extracted JSON is updated.
Select Next to go to the Verify Mappings step. A Mapping Fields... message appears while AgentFlow Unstructured maps the extracted JSON to the data model of your target tenant.
1. Use File View or JSON View to inspect the source content and extracted structure.
2. Use Search to find specific values, if needed.
3. In the Verify entities section on the right pane, review each entity card. Confirm the entity type, mapped attribute count, and confidence value for each entity.
4. Select Verify for an entity type to review the mapping details and confirm that the extracted document field is assigned correctly.
5. Check the values in RELTIO ATTRIBUTES and EXTRACTED FIELD KEY to make sure the field is mapped to the correct Reltio attribute.
6. Review the VALUES column to make sure the extracted content is correct and complete.
7. Review the CONFIDENCE column to understand how certain the system is about that mapped value.
8. Expand rows to review additional mapped content, if needed.
9. Update the selected Reltio attribute for a field if the mapping needs correction.
10. Select Verify attributes to indicate you've verified all attributes and their mappings.
Select Next to continue to the publishing step.
1. Review the template details and select Publish to publish the template.

Result

The template is published and can be selected when you create or configure an AgentFlow Unstructured document pipeline. For more information, see Configure document sources for AgentFlow Unstructured and Set up automated pipelines.

Unify and manage your data

Create and publish an extraction template

Footer