What it does

AWS AI Stack is a monorepo boilerplate for building AI applications on AWS Lambda, API Gateway, DynamoDB, and EventBridge. It includes a vanilla React frontend, JWT-based authentication, an AI chat service with streaming responses via Bedrock, and a business-logic API placeholder. Everything deploys through the Serverless Framework with shared configuration across environments.

The interesting bit

The project uses Lambda Function URLs instead of API Gateway for the AI chat service specifically to support streaming responses—a pragmatic workaround for a genuine AWS limitation. It also structures the auth layer as a shared NPM package consumed by other services via relative paths, which is the kind of “boring but correct” decision that saves headaches later.

Key highlights

• Supports multiple Bedrock models (Claude 3.5 Sonnet, Llama 3.1, Mistral Large 2) with data staying inside AWS
• Domain-oriented architecture: remove the AI chat, auth, or other pieces without breaking the rest
• Serverless Compose ties services together with CloudFormation outputs passed between them
• GitHub Actions for prod deploys and PR preview environments with automatic cleanup
• Custom domain setup via Route53 and ACM for production stages

Caveats

• AWS Bedrock models require manual console approval; new AWS accounts or those with minimal billing history may need to contact AWS support
• The sharedTokenSecret defaults to the literal string "DEFAULT" in non-prod environments, which is documented but easy to miss
• Frontend local development requires manually exporting deployed service URLs as environment variables

Verdict

Worth a look if you want a production-shaped AWS serverless foundation without building auth, streaming, and multi-service deployment from scratch. Skip it if you’re not committed to the AWS ecosystem or need a framework-specific frontend.