Hello Zapier Community!
Based on a requirement for production grade generative AI automations I’ve built an integration to allow humans to review AI generated content in between automations.
For example take a Zapier automation that generates a personalized email using OpenAI. With Human in The Loop, a human can review to make sure the content is correct or fine tune before sending it to the customer.
How it works
To setup a workflow similar to above leveraging human in the loop extension you will have to setup two automations. One to generate the content for the email and send to approval, another one to send the email to customer after approving. It’s straightforward to setup.
Step 1: Setup the action to generate content and send to a human
Breakdown of the flow,
- The trigger listens to new google form submissions.
- Information from the form is then sent to OpenAI to craft an email response.
- Create a new task for a human to review. This action will return a unique link to share with the reviewer.
- Link is emailed to the reviewer.
As part of the configuration I am sending the AI generated content with some metadata,
The output from the new Human in the Loop task is a URL. A human can visit this url in a browser to review the content.
The output can be emailed or sent through Slack to a reviewer, or even create a task in Clickup as a todo.
This is what reviewer will see when visiting the url. The content is editable. Upon approval or rejection, Zapier will trigger ‘Human in the Loop Task Reviewed. This is the second part of the setup
Step 2: Setup the trigger to take action after a human review the content
To handle reviewed content we’ll need to create another automation. For the example shared in the beginning I’ve setup it up as below,
- Using “Human in Loop Task Reviewed” trigger from EnforcedFlow
- A filter condition to make sure only approved ones are processed.
- Send the email to customer
Configuring the trigger
As you can see in below screenshot, the trigger receive the information from first automation upon a review is complete. The trigger receives below information,
- An ID
- Title: The title used to create the task
- Original content: The content used in task creation
- Approved content: The content after approval. If the reviewer doesn’t change anything this will be same as original content
- Review outcome: Outcome of the review, approved or rejected
- Metadata: List of metadata, you can see the email passed in first automation has been passed
Get started
Thanks for staying till the end The extension is still invite only and free to use, you can access it below
https://zapier.com/developer/public-invite/215376/f00e6e4e83c5990562e070d50d5c05a1/