AWS has announced a new feature that will let companies connect their own data sources to foundation models (FMs), which are general AI models that are trained on a large set of data and then can be adapted further for specific uses.  

This is an extension of the company’s recent announcement for Amazon Bedrock that allowed developers to create managed agents using generative AI. Amazon Bedrock is a solution for building generative AI applications using FMs.

According to Amazon, giving these FMs access to your company’s data can help in generating “more relevant, context-specific, and accurate responses.”

The agent searches for the appropriate knowledge bases, retrieves relevant information, and adds that information back into the input prompt to add additional context. It also provides source attribution to whatever data it retrieves for transparency and to prevent hallucinations, which is when an AI model outputs something not based on or related to the input data. 

AWS says that creating these knowledge bases is relatively simple. Users specify the location of the data, select an embedding model, and provide details for the vector database. 

To show its application in the real world, the company gave an example of a tax consulting company that wants to provide its customers with a chatbot that can answer questions about filing taxes. The first step would be to create a knowledge base of tax documents, then configure a Bedrock agent to have access to that knowledge base and then integrate that agent into the chatbot. 

For more information, and for a tutorial on getting started, visit Amazon’s blog post announcing this new feature.