Answer: Use OpenSearch Serverless with properly sized capacity units and index refresh intervals

## Explanation **Correct Answer: C** - Use OpenSearch Serverless with properly sized capacity units and index refresh intervals **Why this is correct:** 1. **Bedrock Knowledge Base Architecture**: Amazon Bedrock Knowledge Bases use vector databases (like OpenSearch Serverless) to store and retrieve embeddings efficiently. When query performance is slow, the vector database configuration is typically the bottleneck. 2. **OpenSearch Serverless Optimization**: Properly sizing capacity units ensures adequate compute resources for query processing, while optimizing index refresh intervals affects how frequently new data becomes searchable and impacts query performance. 3. **Direct Impact on Retrieval Speed**: Unlike the other options, this directly addresses the underlying infrastructure responsible for storing and querying embeddings. **Why other options are incorrect:** - **A) Reduce model temperature**: Model temperature affects the randomness/creativity of text generation, not retrieval speed from a knowledge base. - **B) Use a smaller embedding dimension size**: While smaller embeddings might theoretically be faster, this would require re-embedding all data and could reduce semantic accuracy. It's not the most effective or practical solution. - **D) Add more S3 buckets**: S3 buckets store source documents, not embeddings. Adding more buckets doesn't affect retrieval speed from the vector database. **Key Takeaway**: When experiencing slow query performance with Bedrock Knowledge Bases, focus on optimizing the vector database configuration (OpenSearch Serverless) rather than model parameters or storage configurations.

Author: Ritesh Yadav