A company is migrating a Retrieval-Augmented Generation application to AWS. Documents have already been moved into an S3 bucket. The application requires semantic text search over those files. Which AWS solution meets this requirement?

Amazon Kendra is a highly accurate intelligent search service powered by machine learning, designed for semantic search over various document types, including those in S3. It natively supports S3 connectors for ingesting documents and automatically handles embedding generation and semantic search capabilities. Running an AWS Batch job to compute embeddings and storing them in AWS Glue would require significant custom development for both embedding generation and semantic search implementation, which Kendra provides out-of-the-box. Similarly, using a custom SageMaker notebook and SageMaker Feature Store would involve building a custom search solution, which is more complex than leveraging Kendra's managed service. Amazon Textract is primarily for optical character recognition (OCR) and extracting text and data from documents, not for semantic search capabilities. While Textract might be a precursor for some documents before Kendra, it doesn't perform semantic search itself.