Amazon Bedrock Knowledge Bases now supports advanced parsing, chunking, and query reformulation giving greater control of accuracy in RAG based applications

This article discusses new features in Amazon Bedrock Knowledge Bases that improve the accuracy of responses in retrieval-augmented generation (RAG) applications. These features include:

• Advanced parsing techniques using foundation models (FMs) to handle complex documents with nested tables or images
• Advanced data chunking options:
  • Semantic chunking: Groups related content based on semantic meaning
  • Hierarchical chunking: Organizes data into a hierarchical structure for efficient retrieval
• Custom processing using AWS Lambda functions for custom chunking or metadata processing logic
• Metadata customization for CSV files to separate content and metadata fields
• Query reformulation to break down complex queries into multiple sub-queries for more targeted retrieval

The conclusion highlights how these features give greater control over accuracy and customization for RAG workflows in Amazon Bedrock Knowledge Bases.