AWS AI Lab Nova Sonic and Bedrock

This lab deploys a real-time speech-to-speech AnyTelco support agent where a fictional customer calls in for help. It uses Amazon Nova Sonic in Amazon Bedrock for natural conversation, with S3/CloudFront and Cognito for frontend/auth. A backend on ECS/NLB streams audio and invokes MCP tools (e.g., DynamoDB customer profiles, Bedrock knowledge bases). All infrastructure is provisioned via AWS CDK.

• You will be a mock customer of a telecom company called AnyTelco.
• Customer calls in about upgrading a subscription plan.
• Nova Sonic handles natural dialog, retrieves account details from a mock database, queries a Amazon Bedrock knowledge base via RAG, and adapts responses dynamically.
• Web UI shows transcription, speech output, sentiment over time, and agent tips.
• Conversation shows improved sentiment and natural interruption handling.

• Interact with AnyTelco voice agent to experience working with voice capable models in AWS Bedrock.

- Amazon Bedrock – Nova Sonic (model access required; currently us-east-1 only in this sample).

- Amazon Cognito (auth, user pool + app client).

- Amazon ECS (backend containers) behind NLB (WebSocket).

- Amazon S3 + CloudFront (host/serve the frontend).

- Amazon Bedrock Knowledge Bases (optional: provide KNOWLEDGE_BASE_ID).

- Amazon DynamoDB (optional: customer profile table keyed by phone_number; sample CSV provided for telco).

- Amazon VPC (CDK can create one or you can supply VPC_ID).

- AWS CDK/CloudFormation (end-to-end infra provisioning).