v0.3.0

[0.3.0]

Added

• Visual Edit Feature for Document Processing

  • Interactive visual interface for editing extracted document data combining document image display with overlay annotations and form-based editing.
  • Split-Pane Layout, showing page image(s) and extraction inference results side by side
  • Zoom & Pan Controls for page image
  • Bounding Box Overlay System (Pattern-1 BDA only)
  • Confidence Scores (Pattern-1 BDA only)
  • User Experience Benefits
    • Visual context showing exactly where data was extracted from in original documents
    • Precision editing with visual verification ensuring accuracy of extracted data
    • Real-time visual connection between form fields and document locations
    • Efficient workflow eliminating context switching between viewing and editing

• Enhanced Few Shot Example Support in Pattern-2

  • Added comprehensive few shot learning capabilities to improve classification and extraction accuracy
  • Support for example-based prompting with concrete document examples and expected outputs
  • Configuration of few shot examples through document class definitions with examples field
  • Each example includes name, classPrompt, attributesPrompt, and imagePath parameters
  • Enhanced imagePath Support: Now supports single files, local directories, or S3 prefixes with multiple images
    • Automatic discovery of all image files with supported extensions (.jpg, .jpeg, .png, .gif, .bmp, .tiff, .tif, .webp)
    • Images sorted alphabetically in prompt by filename for consistent ordering
  • Automatic integration of examples into classification and extraction prompts via {FEW_SHOT_EXAMPLES} placeholder
  • Demonstrated in config_library/pattern-2/few_shot_example configuration with letter, email, and multi-page bank-statement examples
  • Environment variable support for path resolution (CONFIGURATION_BUCKET and ROOT_DIR)
  • Updated documentation in classification and extraction README files and Pattern-2 few-shot examples guide

• Bedrock Prompt Caching Support

  • Added support for <<CACHEPOINT>> delimiter in prompts to enable Bedrock prompt caching
  • Prompts can now be split into static (cacheable) and dynamic sections for improved performance and cost optimization
  • Available in classification, extraction, and summarization prompts across all patterns
  • Automatic detection and processing of cache point delimiters in BedrockClient

• Configuration Library Support

  • Added config_library/ directory with pre-built configuration templates for all patterns
  • Configuration now loaded from S3 URIs instead of being defined inline in CloudFormation templates
  • Support for multiple configuration presets per pattern (e.g., default, checkboxed_attributes_extraction, medical_records_summarization, few_shot_example)
  • New ConfigurationDefaultS3Uri parameter allows specifying custom S3 configuration sources
  • Enhanced configuration management with separation of infrastructure and business logic

Fixed

• Lambda Configuration Reload Issue
  • Fixed lambda functions loading configuration globally which prevented configuration updates from being picked up during warm starts

Changed

• Simplified Model Configuration Architecture

  • Removed individual model parameters from main template: Pattern1SummarizationModel, Pattern2ClassificationModel, Pattern2ExtractionModel, Pattern2SummarizationModel, Pattern3ExtractionModel, Pattern3SummarizationModel, EvaluationLLMModelId
  • Model selection now handled through enum constraints in UpdateSchemaConfig sections within each pattern template
  • Added centralized IsSummarizationEnabled parameter (true|false) to control summarization functionality across all patterns
  • Updated all pattern templates to use new boolean parameter instead of checking if model is "DISABLED"
  • Refactored IsSummarizationEnabled conditions in all pattern templates to use the new parameter
  • Maintained backward compatibility while significantly reducing parameter complexity

• Documentation Restructure

  • Simplified and condensed README
  • Added new ./docs folder with detailed documentation
  • New Contribution Guidelines
  • GitHub Issue Templates
  • Added documentation clarifying the separation between GenAIIDP solution issues and underlying AWS service concerns