Question

A legal research firm uses an Amazon Bedrock Knowledge Base backed by Amazon OpenSearch Serverless as its vector store. The knowledge base ingests thousands of case documents, each tagged with attributes such as jurisdiction, practice_area, and document_year. Attorneys frequently complain that RAG responses mix in irrelevant cases from the wrong jurisdiction. The team wants to constrain retrieval to only the jurisdiction relevant to each query while keeping a single knowledge base. What is the most effective approach?

Accepted Answer

Include the jurisdiction, practice_area, and document_year as metadata attributes during ingestion, then apply metadata filters on the RetrieveAndGenerate API call at query time.

Answer

Create a separate Bedrock Knowledge Base per jurisdiction and route queries to the correct one in application code.

Answer

Increase the number of retrieved chunks (topK) so more documents are returned and let the foundation model ignore irrelevant jurisdictions.

Answer

Embed the jurisdiction name into every chunk of text before ingestion so the embeddings encode jurisdiction similarity.