Diagram 10 of 12

Generative AI with RAG on Amazon Bedrock

An assistant that answers from your own documents. Lambda searches OpenSearch for the relevant chunks and sends them to an Amazon Bedrock model together with the question. Documents are ingested from S3.

Amazon API GatewayAWS LambdaAmazon BedrockAmazon OpenSearch ServerlessAmazon S3

Diagram

Comments

The model does not know your documents. RAG adds the context retrieved from OpenSearch before generating the answer.
Embeddings must be generated with the same model during ingestion and querying. If you change models, you must reindex.
Review Bedrock's data handling policies and encryption options before loading sensitive information.
For repeated questions, a cache in DynamoDB reduces cost and latency.

Information flow

Amazon API Gateway
The user sends the question to API Gateway.
AWS Lambda
API Gateway invokes the Lambda function that orchestrates the RAG flow.
Amazon OpenSearch Serverless
Lambda searches the vector index for the most similar document chunks.
Amazon Bedrock
Lambda sends the question and the retrieved context to the Bedrock model.
Amazon Bedrock + API Gateway
Bedrock returns the answer, and the result reaches the user through API Gateway.