AI-ASSISTED AWS INFRASTRUCTURE CODE GENERATION AND SECURE DEPLOYMENT PIPELINE

Writing infrastructure as code by hand demands detailed knowledge of cloud-service syntax and security practice, and AI coding assistants that generate such code produce drafts that cannot be trusted without verification. This project presents an AI-assisted AWS infrastructure code generation and secure deployment pipeline that places a controlled, event-driven gate between a generated template and a live deployment. CloudFormation templates, drafted with Amazon Q Developer, are uploaded to an Amazon S3 bucket. An S3 object-created event invokes an AWS Lambda processor, which retrieves the template and validates it with AWS CloudFormation. Every outcome is written back to S3 as a JSON artifact. Two processing paths are distinguished by key prefix: templates under templates/ are validated only, while templates explicitly placed under deploy/ are validated and then deployed with CloudFormation create_stack using a dedicated service role. Three separate IAM identities – processor, deployment and application – keep provisioning permissions out of the processor’s runtime role. The pipeline was tested in the AWS Mumbai region. Valid templates produced validation results without creating infrastructure, a malformed template was rejected with an error result, and a deploy-path upload created a twelve-resource Notes API stack (API Gateway, Lambda, DynamoDB, IAM) that reached CREATE_COMPLETE in about 46 seconds. HTTP requests to the deployed endpoint created, retrieved and listed notes stored in DynamoDB. Troubleshooting during testing led to permissions being added to the deployment role only when a failure demonstrated the need. The result is a small, serverless and reusable pattern that shortens the path from an architecture idea to checked, deployable infrastructure while keeping deployment an explicit decision. Keywords: infrastructure as code, generative AI, Amazon Q Developer, AWS Lambda, CloudFormation, serverless, event-driven automation, IAM.