Knowledge Bases for Amazon Bedrock delivers a fully managed Retrieval Augmented Generation (RAG) experience, empowering you to securely connect your foundation models to your company’s data sources. By augmenting the model’s responses with relevant information retrieved from your knowledge base, you can generate more accurate, context-specific, and transparent outputs, without the need for continuous retraining.
One of the key features of Knowledge Bases is the new RetrieveAndGenerate API. This powerful API allows you to directly retrieve relevant information from your knowledge base and have Amazon Bedrock generate a response from the results, all within a single API call. By specifying the foundation model you want to use, you can seamlessly integrate RAG into your applications, unlocking the full potential of your company’s data.
Imagine you work at an investment firm and want to fine-tune the Cohere Command Light model to summarise financial reports and analyst commentary. By providing your own dataset of financial documents, you can train the model to understand industry-specific terminology and generate concise, relevant summaries that your analysts can leverage to make informed decisions. You can then further enhance the model’s capabilities by connecting it to your company’s knowledge base, allowing it to retrieve and incorporate relevant information from your internal data sources, such as market research reports and client portfolios.
Similarly, if you work at an e-commerce company and want to fine-tune the Amazon Titan Text model to be more knowledgeable about your product catalogue, customer service interactions, and industry trends, you can leverage the power of Knowledge Bases. By providing a dataset of your company’s internal documents, such as shareholder letters, product descriptions, and customer support transcripts, you can train the model to understand the nuances of your business and generate more relevant and contextual responses, further enriched by the information retrieved from your knowledge base.
One of the key advantages of Amazon Bedrock’s model customisation and knowledge base capabilities is the emphasis on data privacy and network security. Your data, including prompts, completions, custom models, and data used for fine-tuning, continued pre-training, or populating your knowledge base, remains private to your AWS account and is not used for service improvement or shared with third-party model providers. Additionally, all data is encrypted in transit and at rest, and you can use AWS PrivateLink to create a private connection between your VPC and Amazon Bedrock.
As you explore the possibilities of customising and enriching your models with Amazon Bedrock, it’s important to understand the billing considerations. The service charges for model customisation, storage, and inference, with model customisation billed per tokens processed, model storage billed per month per model, and inference billed hourly per model unit using provisioned throughput.