Building a Multi-Modal RAG pipeline with AWS CDK,Strands,AWS Stepfunctions

References

https://docs.aws.amazon.com/lambda/latest/dg/with-sqs.html#events-sqs-queueconfig

Prerequisites

• AWS CLI
• AWS Bedrock Access AWS Credentials
• Strands Agent CLI
• Docker
• python >=3.11

Create CDK PROJECT

cdk init app --language=python

Import And Configure Dependencies

aws-cdk.aws-lambda-python-alpha==2.175.1a0
cdklabs.generative-ai-cdk-constructs==0.1.309
pre-commit==4.1.0
ruff==0.9.1

lambda_fns

aws-lambda-powertools[tracer]
boto3>=1.34.0
strands-agents>=0.1.0
strands-agents-tools>=0.1.0

Create S3 bucket

• done

Create 4 Lambda Function Skeletons

• done

• upload_processor.py ---> uploads/(s3)

• queue_processor.py --> polls from the sqs queue

• save_textract_text_function.py --> saves text extracted from pdf files

• save_transcribed_text_function.py--> saves transcribed text from audio/video

Create 4 Lambda function resources

• done

Create SQS and DLQ

• done

Grant Appropriate Permissions to S3/Lambda/SQS

• done

Create PDF Processor State Machine Resource

• done

Create PDF Statemachine Workflow

• done

Grant Appropriate Permissions to PDF State Machine

• done

Create Audio/Video Processor State Machine Resource

• done

Create Audio/Video Statemachine

• done

Grant Appropriate Permissions to Audio/Video State machine

• done

Update Lambda functions(extract text) with AI Agent With Strands

Create AWS Bedrock Knowledge base with Custom Datasource and Pinecone

• done

Grant extract text lambda function to save to KB

• done

Conclusion

• create an endpoint that enables users to query this KB.