AWS re:Invent 2025 - Build Enterprise AI Apps Faster: Amazon Bedrock's Multimodal Solutions -AIM3341

Accelerating Multimodal Content Processing with Amazon Bedrock Data Automation

Multimodal Content Challenges

• 80% of enterprise content is unstructured, spanning documents, images, videos, and audio
• Extracting insights from this diverse, high-volume content is incredibly difficult and time-consuming
• Key challenges include:
  • Handling the wide variety of content formats and schemas
  • Achieving the necessary accuracy and scale for production use cases
  • Maintaining auditability and transparency of the extraction process
  • Integrating and orchestrating multiple specialized services and tools

Amazon Bedrock Data Automation (BDA)

• A unified API that allows processing of images, documents, video, and audio through a single interface
• Simplifies implementation by handling the orchestration, model selection, and transformations
• Enables customization of output schemas and normalizations to fit downstream systems
• Provides confidence scores, grounding, and auditability to support human review and governance

Key BDA Use Cases

1. Intelligent Document Processing: Extracting insights from insurance claims, loan applications, and other document-heavy workflows
2. Intelligent Search and Analytics: Powering search, categorization, and analysis of call center transcripts, media assets, and other multimodal content
3. Media Analysis and Content Discovery: Enabling search, summarization, and metadata extraction for large video and image libraries
4. Agentic Workflows: Integrating multimodal content processing directly into agent-facing applications and processes

BDA Architecture and Capabilities

• Provides both standard and custom output configurations
• Standard output offers pre-defined schemas and extraction capabilities per modality
• Custom output allows defining tailored schemas and extraction rules using natural language instructions
• Handles the underlying orchestration, model selection, and transformations transparently

Air's Journey with BDA

• Air is a creative operations platform managing 8+ PB of multimodal content for 100,000+ users
• Previously relied on a fragmented pipeline of specialized services, which was complex to maintain and scale
• Adopted BDA to unify their multimodal content processing with a single API
• Benefits include:
  • Simplified implementation and reduced maintenance overhead
  • Improved security by keeping data within AWS infrastructure
  • Cost-effective scalability to handle hundreds of thousands of daily content uploads
  • Ability to customize output schemas to match their user experience requirements

What's New and Coming Soon with BDA

• Expanded language support for document processing (now 5 languages)
• Increased image processing speed and added support for more video/audio formats
• Upcoming features:
  • Document accuracy optimization using ground truth labeling
  • Entity detection with custom pronunciation and recognition models
  • Expanded regional coverage (15+ AWS regions by end of 2023)

Key Takeaways

• Multimodal content processing is a significant challenge for enterprises, with 80% of data being unstructured
• Amazon Bedrock Data Automation provides a unified API to simplify the extraction of insights from diverse content types
• BDA handles the underlying orchestration, model selection, and transformations, enabling greater customization and auditability
• Customers like Air have seen major benefits in terms of reduced complexity, improved security, and cost-effective scalability
• BDA is continuously expanding its capabilities, with upcoming improvements in accuracy, entity detection, and regional coverage