Generative AI Serverless - RAG using Bedrock Converse API & Lambda - Chat with your document

In this video, I am going to show you how to build a serverless GenAI RAG solution to implement a document chat feature using Amazon Bedrock Converse API and Lambda.
While Bedrock already provided a 'Chat with your Document' feature, with the introduction of Converse API, a DocumentBlock has been added so that user can provide a document for contextual conversation and can ask questions based on the data in this document. Document can be hosted in S3 bucket or can be provided to function at runtime.
Use of DocumentBlock in Amazon Bedrock's Converse API can be considered a form of RAG for generative AI since it leverage the data provided in the document to enhance the generation of contextual responses.
Bedrock Knowledge Base : • Generative AI Serverle...
Lambda, SAM: • How to develop AWS Lam...

[@DarrenMariadas]

what is the differentiating use case as to when to implement the RetrieveAndGenerate API vs the Converse API when building a RAG Chatbot?

[@CloudWithGirish]

Darren, Converse API is a newer addition and is becoming my first preference when using Bedrock. Converse API provides a consistent interface to use multiple models. if use case requires getting the source/citation of the KB document used in the RAG solution then, I will use RetrieveAndGenerate API.

[@siddartha5407]

Can we use bedrock embiddings with converse API from vector database?

[@CloudWithGirish]

Currently, embedding models not supported by Converse API however, you can use Bedrock InvokeModel.

[@siddartha5407]

@@CloudWithGirish Sure, Thank you for the reply.

[@suneja2110]

Can we use converse api if we just need to summarize documents like pdf and images without any conversation (single turn)? Also do we know what is retention period for the uploaded document in aws?

[@CloudWithGirish]

embedding and image not supported however you can use for summarization and PDF is supported. Also, if you use console for bedrock/KB, retention is for current session only. if you use S3 for the document, you can control the retention and remove the document via event, life cycle policy or via console.