AI Text Extraction Utility (Llama)

Workflow

Artificial Intelligence

Universal

Beginner

Use Template

This is a 'Workflow' template which means that it is a single standalone workflow.

Some workflow templates can be modified to work with other workflow templates - e.g. to convert a data sync between two services from uni-directional to bi-directional

Overview

Copy

Use this flexible AI Utility Callable to start infusing your automations and business processes with AI.

This utility workflow can extract structured fields of information from unstructured text and data.

The only setup required is your AWS Bedrock credentials.

Prerequisites

Copy

To implement this AI powered automation, you will need:

• API credentials for AWS Bedrock

Note: while this automation uses script steps, you do not need to understand or write scripts to benefit. If you want to learn more about what the script steps do we included comments in the code so you could reverse engineer what is happening if you want to.

Getting Live

Copy

Mandatory steps:

1. Map the source text that you want to extract data from

2. Define the field(s) that you want AI to pull from the source text

Optional steps (to refine your extraction job):

• Add words that you want the AI job to ignore (these words will not be extracted)

• Add additional instructions that will be appended to the system prompt

• Include example Source Text and Extraction Answers (very effective for complex extractions)

Example Use Cases

Copy

• Scrubbing Autoresponder Emails (level up: use in coordination with a classification AI utility for best in class marketing ops)

• Identifying PII. Before storing information or forwarding it downstream in an automation you can identify if any text contains PII and define what counts as PII for your use case. Excellent for success teams scaling out public facing knowledge bases from real customer tickets.

• Getting the Company information when many companies are mentioned. Let AI figure out what the primary company is in a body of text (i.e. a phone call transcription) when many company names are mentioned it can use semantic understanding to pick out the correct one and not the vendors or affiliates.

• and so many more.

Here are a few examples of implementation (with source text hardcoded for visibility):