Worked on the aws-samples/sagemaker-genai-hosting-examples repository to implement a new asynchronous inference feature for SageMaker, enabling inline payloads in InvokeEndpointAsync requests. This enhancement allows users to send request bodies directly in the API call, supporting payloads up to 128KB and removing the need for S3 uploads. The solution was developed using Python and leveraged AWS services, particularly SageMaker and boto3, to ensure seamless integration. An end-to-end deployment, invocation, and validation workflow was documented in a Jupyter notebook, demonstrating the feature’s usage with Hugging Face PyTorch inference containers and providing clear guidance for reproducibility and validation.