Create a SageMaker Transformer object from a trained model with transformer = model.transformer(instance_count=2, instance_type='ml.m5.xlarge', output_path='s3://...')
Start the job with transformer.transform(data='s3://input-path/', content_type='text/csv', split_type='Line') to process CSV files line by line
Monitor job status with transformer.wait() for blocking execution or poll boto3 sagemaker_client.describe_transform_job() for async workflows
Read output files from the output S3 path — each input file produces a corresponding .out file with model predictions
Tune throughput with max_concurrent_transforms and max_payload parameters to balance speed against memory pressure on the instance
Known gotchas
Batch Transform does not support real-time autoscaling — you must provision enough instance_count upfront to handle the full dataset within your time budget
The split_type='Line' setting is essential for CSV data — omitting it sends the entire file as a single request and most containers will reject it or return a single aggregated prediction
Batch Transform creates a new model endpoint internally for the job duration — ensure the SageMaker execution role has permission to create endpoints in addition to transform job permissions
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp