With the 26R1 release, you can enable the Intake Agent in Vaults using Veeva AI for Safety. The Intake Agent processes unstructured and semi-structured Adverse Event Report (AER) data from sources such as CIOMS forms, MedWatch forms, and emails. The agent extracts case information from source documents and automates the initial states of intake by mapping the source data to structured fields in Vault records, allowing for a more efficient triaging process.

The Intake Agent uses a Large Language Model (LLM) to identify and extract patient, reporter, country, product, and event details. When the agent identifies a potential ICSR in a file, Vault copies the extracted details to the generated Inbox Item and adds document annotations to highlight the specific text used for the extracted field values. Vault also flags any data anomalies for review. Users can review the extracted data and make any necessary updates on the Document Intake Highlighter panel.

When enabled, the Intake Agent runs in the following scenarios for documents associated with applicable Transmission Profiles:

Prerequisites

Before configuring this feature in your Vault, you must complete the following:

Update Transmission Profile Layouts

Add the Enable Intake Agent field to the Transmission Profile object layouts for the Inbound Email and Partner Exchange Profile object types so that you can define this value on the applicable records.

Update Transmission Profiles

To allow the Intake Agent to run on supported documents associated with specific Transmission Profiles, update the Enable Intake Agent field value to Yes on applicable records.

Configure the Extract ICSR with Intake Agent Document Action

To allow users to manually run the Intake Agent, configure the Extract ICSR with Intake Agent user action on applicable states of the AER Lifecycle. Users can run the action on Adverse Event Report classified documents not yet associated with an Inbox Item. This allows users to run the agent on documents uploaded prior to enabling this feature and on any documents on which the agent previously ran but detected errors.