Skip to content

Get started with assisted document suggestions

Deploy the regional model package in your own AWS account, send the included fictional example and remove temporary compute when finished.

Subscribe and choose the package

  1. Open the Marketplace listing. Review the scope, current version, software price, trial, EULA and refund policy.
  2. Subscribe, choose Continue to configuration and select the region. The notebook and measured workload use us-east-1.
  3. Copy the subscribed regional model-package ARN. Use this product's package, not another model or a seller's private test ARN.
  4. Provide an existing SageMaker execution role from your account. Your caller needs permission to orchestrate the chosen resources; the execution role separately grants SageMaker access to the required private data.

Open the complete notebook

Use the Boto3 notebook. Its default Run All is offline and makes no AWS requests. It contains fictional input text, so no dataset search or customer upload is required. Use a current boto3 release supporting conditional S3 PutObject.

Supply your regional package through DOCUMENT_CLASSIFIER_MODEL_PACKAGE_ARN and your execution role through SAGEMAKER_EXECUTION_ROLE_ARN. Review costs and cleanup, then explicitly enable RUN_DEMO. The notebook creates one temporary ml.c5.xlarge endpoint with network isolation, invokes it and removes its own endpoint, configuration and model in finally.

Keep the printed resource name

Kernel termination or a client timeout does not guarantee cleanup. Inspect the exact endpoint in SageMaker and delete it if it remains. Charges continue while it exists.

Optional batch example

Set DOCUMENT_CLASSIFIER_BATCH_BUCKET to an existing authorized us-east-1 bucket. Your execution role needs read/write access to the unique example prefix. Explicitly enable RUN_BATCH_DEMO after reviewing the settings.

The notebook includes three fictional ES/PT/EN records, uses one C5 instance with SingleRecord and concurrency one, and removes its model after the job stops. It creates no bucket or role. Small task-owned input/output S3 objects are retained for inspection; apply your own retention policy to those exact objects.

The seller's own certified source uses a separately verified Catalog association path. That seller rehearsal does not establish an independent buyer checkout or purchase.

Next: input/output contract and batch settings.

Last updated November 16, 2026 · Suggest an improvement