Separating knowledge in RAG solutions

Yes, there is a solution to separate knowledge for each application while using a single knowledge base in your RAG setup with Amazon Bedrock. Instead of creating multiple knowledge bases, you can leverage metadata or tags to differentiate between documents belonging to different applications.

When uploading documents to your S3 bucket, you can include application identifiers as object metadata or tags. This approach allows you to use a single knowledge base for all applications while still maintaining separation.

When querying the knowledge base using the Retrieve or RetrieveAndGenerate API, you can then filter the results based on these metadata or tags to retrieve documents specific to a particular application. This filtering capability is supported in the API calls, allowing you to retrieve only the relevant information for each application.

For example, when uploading documents to S3, you could add metadata or tags that identify which application each document belongs to. Then, when using Lex with different intents for different applications, you can include filters in your API calls to the knowledge base based on these metadata or tags. This way, each intent will only retrieve and use the documents relevant to its specific application.

This approach allows you to maintain the simplicity and cost-efficiency of using a single knowledge base while still providing the necessary separation of knowledge for your multiple applications. It offers a good balance between isolation of application-specific information and the management simplicity of a unified knowledge base.
Sources
Amazon Bedrock Knowledge Base -Limit of 50 Knowledge Base | AWS re:Post
Knowledge bases for Amazon Bedrock - AWS Prescriptive Guidance