Question

A healthcare company is building a patient-support chatbot using Amazon Bedrock Knowledge Bases with an Amazon S3 data source containing clinical documents. Compliance requires that all patient PII (names, medical record numbers) be removed before any text is embedded and stored in the vector database, and that all data remains within the eu-west-1 region. The team wants a scalable, managed approach with minimal custom infrastructure. Which solution best meets these requirements?

Accepted Answer

Run an Amazon Comprehend PII detection and redaction job on the source documents in S3, store the redacted output in an eu-west-1 bucket, and configure the Bedrock Knowledge Base in eu-west-1 to ingest from that bucket.

Answer

Configure the Bedrock Knowledge Base to ingest the raw documents directly and rely on a guardrail applied at inference time to block PII from appearing in chatbot responses.

Answer

Enable Amazon Macie on the source bucket to automatically redact PII in place, then point the Bedrock Knowledge Base at the same bucket.

Answer

Export all documents to a global Amazon OpenSearch Serverless collection, use a Lambda function to strip PII during query time, and store embeddings without region restrictions.