Newsroom Agentic #AWS #Multi-Agent System #Document Classification #AI Model #Amazon Bedrock

AWS Releases Multi-Agent Document Classification System: Combining Textual and Visual Analysis for Enhanced Accuracy

By Mr.Xu

Published: 2026-08-18 · 2 views

中文阅读 (Chinese) English Version

Summary:AWS has launched a multi-agent document classification system that leverages Amazon Bedrock AI models and the Strands Agents SDK to enable efficient collaboration between textual and visual analysis. The system utilizes Anthropic's Claude Haiku 4.5 for advanced textual reasoning and Amazon Titan Multimodal Embeddings for visual similarity search, significantly improving the accuracy of insurance document classification. Test results demonstrate that the system achieves 100% accuracy in handling

AWS Releases Multi-Agent Document Classification System

AWS has launched an innovative multi-agent document classification system designed to address the limitations of traditional automated classification methods in handling complex documents. The system achieves efficient collaboration through the following core components:

Core Components

Document Analysis Agent: Powered by Anthropic's Claude Haiku 4.5, this agent performs advanced textual reasoning and legal language parsing on document content.
Vector Similarity Search Agent: Utilizing Amazon Titan Multimodal Embeddings, this agent converts documents into high-dimensional vector representations for visual similarity search.
Validation Agent: Acting as the coordinator, this agent uses the Strands Agents SDK to enable collaboration between agents, ensuring the accuracy and consistency of classification results through cross-validation and confidence scoring.

Technical Highlights

Multi-Agent Collaboration: By decomposing tasks into specialized domains, agents can focus on their areas of expertise, thereby improving overall classification accuracy.
Combination of Textual and Visual Analysis: Claude Haiku 4.5 excels at understanding complex document content, while Amazon Titan Multimodal Embeddings focuses on capturing the visual and structural features of documents, enabling a more comprehensive analysis.
Scalability and Modularity: Agents can be developed and improved independently, and the system architecture is flexible and easy to extend.

Performance

In testing, the system achieved 100% accuracy on a sample of 20 documents, significantly outperforming traditional methods such as Amazon Textract, Amazon Comprehend, and Amazon Bedrock Data Automation (BDA). Specific data is as follows:

| Method | Accuracy | Average Processing Time | Complexity | Cost | |---|---|---|---|---| | Amazon Textract + Keywords | 25% | 2.88s | Low | Low | | Amazon Comprehend + Entity Recognition | 25% | 3.26s | Medium | Low | | Amazon Bedrock Data Automation | 70% | 25.7s | Medium | Medium | | Multi-Agent System | 100% | 23.3s | High | Medium |

Industry Impact

The release of this system marks a significant milestone in the field of AI document processing, particularly in scenarios where accuracy is critical, such as insurance, finance, and legal industries. Its advantages include:

High-Precision Classification: The combination of multi-agent collaboration and textual and visual analysis significantly improves classification accuracy.
Explainability: Agents provide detailed reasoning processes, facilitating auditing and explaining classification results.
Flexibility: The system can be customized and extended according to specific needs, adapting to different application scenarios.

Developer Recommendations

Gradual Implementation: Developers are advised to first implement the Document Analysis Agent and then gradually add the Vector Similarity Search Agent to enhance the processing of visual features.
Adjust Confidence Thresholds: Adjust the confidence thresholds between agents according to the actual application scenario to find the best balance between automation and human review.
Leverage AWS Resources: Make full use of the resources provided by Amazon Bedrock and the Strands Agents SDK to optimize system performance and cost.

Conclusion

AWS's multi-agent document classification system, by combining textual and visual analysis, significantly improves classification accuracy and provides a new solution for the AI document processing field. The system is not only applicable to the insurance industry but also has broad industry application potential.