
Amazon Shopping migrated from self-hosted ML models to Amazon Rekognition Content Moderation API to automatically detect harmful images in product reviews, achieving higher accuracy, reduced costs, and operational efficiency.


<div><p>This article describes how Amazon Shopping migrated from self-hosted ML models to Amazon Rekognition Content Moderation API for reviewing harmful images in product reviews.</p><ul><li>Amazon receives 1.5 billion reviews annually requiring content moderation at scale</li><li>Self-hosted models automated only 40% of image moderation decisions with ongoing complexity challenges</li><li>Amazon Rekognition Content Moderation API replaced self-managed models for nudity and NSFW detection</li><li>Migration resulted in ~1 million additional images automatically moderated annually without human review</li><li>Achieved significant cost savings by reducing human moderator workload and DevOps effort</li><li>Simplified system architecture, freeing engineering time for innovative features</li></ul><p>Migrating to Rekognition's managed service improved accuracy, reduced operational complexity, and enabled cost savings while maintaining content safety standards.</p></div>


Related articles