Improve contract search accuracy with auto-generated filters in Amazon Bedrock

This article describes AIDA, an AI-driven annotation solution that transforms unstructured contracts into searchable intelligence using Amazon Bedrock Knowledge Bases and retrieval-augmented generation (RAG).

• Implements implicit filtering to automatically apply metadata-based conditions before semantic search
• Uses explicit filtering at the application layer to enforce business policies and compliance requirements
• Enriches document chunks with metadata values like governing law, effective date, and contract type
• Combines filtering and metadata enrichment to dramatically improve RAG accuracy and reduce hallucinations
• Tested on Contract Understanding Atticus Dataset showing metadata enrichment improved answer quality from partial to most reliable

AIDA provides enterprise-ready contract intelligence by grounding legal document analysis in proper context and access boundaries, reducing manual review effort while maintaining accuracy for critical business decisions.