This notebook is triggered by a lambda function which has done postprocess works after Textract API. For A2I work directly within your API calls to Textract's Analyze Document API, refer to this official sample notebook
After the solution stack has deployed, you'll still need to set up the following:
- Create a private team in Amazon SageMaker Ground Truth, setting yourself up with credentials as a reviewer (or re-use your existing team if you created one already for the preprocessing module)
- Upload the provided worker task template as a custom task template in Amazon A2I
- Create an A2I Human Review Workflow linking the template and your review 'team'
- In the AWS SSM Parameter Store console, find the deployed stack's
DefaultHumanFlowArnparameter. - Edit your parameter to set the Value as the ARN of your A2I workflow.
We provide a detailed walkthrough of these steps in a2i_humanloop.ipynb.
Note: For more than 70 official custom task UIs, check out https://github.com/aws-samples/amazon-a2i-sample-task-uis
To run the demo, follow the steps below
- Join the humanflow-team private team
- Upload the receipt to the web ui and make sure it triggers the human review
- Login to the private team labelling console https://g27hwd6auf.labeling.us-east-1.sagemaker.aws
- There will be a task named "human flow task"
- Open the task and review/amend the textract outputs and complete the task