Question #686
A retail chain is developing a real-time inventory tracking system for its stores. The company uses ceiling-mounted cameras to monitor shelf stock levels. The company has trained an Amazon SageMaker machine learning (ML) model to detect low stock items from images. The company needs to trigger alerts for restocking via a local notification system, even during internet outages. A Linux server on premises hosts a REST API that manages the alert system.
How should the company deploy the ML model to meet these requirements?
Configure Amazon Kinesis Data Streams to ingest images from the cameras. Use AWS Lambda to extract frames and store them in Amazon S3. Deploy a SageMaker endpoint for the ML model. Trigger a Lambda function to run inference on new images and invoke the local API via AWS Direct Connect when low stock is detected.
Install AWS IoT Greengrass on the local server. Deploy the ML model to the Greengrass core. Create a Greengrass component to capture images from the cameras, perform local inference, and trigger the REST API directly when low stock is identified.
Use AWS Outposts to host a SageMaker endpoint and EC2 instance locally. Capture images from the cameras, run inference on the EC2 instance via the SageMaker endpoint, and configure the instance to call the local API when low stock is detected.
Deploy AWS Panorama Appliances on each camera. Train the ML model directly on the Panorama devices. Use Amazon EventBridge to route low stock alerts to the local API via an on-premises AWS Storage Gateway.
Explanation
Answer B is correct because AWS IoT Greengrass allows deploying the ML model locally on the on-premises server, enabling inference and alert triggering even during internet outages. The Greengrass component can capture images, run local inference using the deployed model, and directly invoke the REST API without relying on cloud connectivity.
Other options fail due to cloud dependencies: Option A uses cloud services (Kinesis, Lambda, SageMaker endpoints) that require internet. Option C relies on AWS Outposts, which still needs connectivity to AWS. Option D uses Panorama devices and EventBridge, which may not handle local API integration robustly during outages. Key points: Edge computing with Greengrass ensures offline operation; local inference and API calls avoid cloud dependencies.
Answer
The correct answer is: B