Enhancing Amazon Bedrock with Knowledge Bases: New Features for RAG Generation

Knowledge Bases for Amazon Bedrock are a powerful tool that allows you to securely connect foundation models (FMs) in Amazon Bedrock to your company data for Retrieval Augmented Generation (RAG). This feature enables you to access additional data to generate more relevant, context-specific, and accurate responses without the need to retrain the FMs. In this blog post, we will explore two new features specific to the RetrieveAndGenerate API: configuring the maximum number of results and creating custom prompts with a knowledge base prompt template.

## Overview and Benefits of New Features
The maximum number of results option gives you control over the number of search results to be retrieved from the vector store and passed to the FM for generating the answer. This customization allows you to provide more or less background information for generation, depending on the complexity of the question. By fetching up to 100 results, you can improve relevance and reduce hallucination in the generated response.

The custom knowledge base prompt template feature allows you to replace the default prompt template with your own to customize the tone, output format, and behavior of the FM when responding to a user’s question. This level of customization enables you to fine-tune terminology, add custom instructions, and examples tailored to your specific workflows.

## How to Use These Features
### Configure the Maximum Number of Results Using the Console
To configure the maximum number of results using the console, follow these steps:
1. Navigate to the Amazon Bedrock console and select Knowledge bases.
2. Choose the knowledge base you want to configure.
3. Select Test knowledge base.
4. Click on the configuration icon.
5. Choose Sync data source before testing.
6. Under Configurations, set the Maximum number of source chunks as needed.

By adjusting the maximum number of results, you can enhance the accuracy of the generated response based on the retrieved information. Different configurations can yield different results, as shown in examples within the post.

### Customize a Knowledge Base Prompt Template Using the Console
To customize the default prompt with your own template, follow these steps on the console:
1. Start testing your knowledge base.
2. Enable Generate responses and select the model for response generation.
3. Choose Apply and edit the Knowledge base prompt template section.

By customizing the prompt template, you can influence the tone, language, and structure of the generated response for your specific use case, as demonstrated in provided examples.

## Conclusion
Knowledge Bases for Amazon Bedrock offer valuable features to enhance RAG-based applications. By utilizing the maximum number of results configuration and custom prompt templates, you can improve the performance and accuracy of generated responses tailored to your needs. These enhancements provide greater flexibility and control, enabling you to deliver customized experiences for your applications.

For more information and resources on implementing these features in your AWS environment, refer to the documentation provided. If you have any questions or need assistance, feel free to reach out to the authors of this post for expert guidance and support in leveraging generative AI solutions.

**About the Authors:**
– Sandeep Singh: Senior Generative AI Data Scientist at Amazon Web Services
– Suyin Wang: AI/ML Specialist Solutions Architect at AWS
– Sherry Ding: Senior AI/ML Specialist Solutions Architect at AWS

Stay tuned for more insights and updates on Amazon Bedrock’s Knowledge Bases and generative AI capabilities. Happy innovating!