infrastructure
Building a Restaurant Voice Ordering System with Amazon Bedrock and Nova Sonic
AWS demonstrates how to create a voice-based restaurant ordering system using Amazon Bedrock AgentCore and Nova 2 Sonic, integrating telephony with backend services for seamless customer interactions.
AS1 NewsSource: aws.amazon.com
This post describes the development of an AI-driven voice ordering system designed for restaurant environments, addressing the challenge of handling high call volumes during peak hours. The system leverages Amazon Bedrock AgentCore to host and run conversational agents, and Amazon Nova 2 Sonic for real-time speech recognition and synthesis. It connects to a restaurant's backend through the Model Context Protocol (MCP), enabling flexible and decoupled architecture.
The architecture comprises three layers: telephony, agent, and backend. The telephony layer uses Amazon Chime SDK Voice Connector to manage inbound calls, which are bridged to the agent layer via a SIP gateway running on Amazon ECS and AWS Fargate. This setup allows audio streams to be translated between RTP packets and WebSocket frames, facilitating communication with the agent.
The agent layer runs each call in a microVM within AgentCore Runtime, ensuring isolation and scalability. Amazon Nova 2 Sonic handles speech processing, recognizing diverse accents and handling interruptions, while the agent interacts with backend services through MCP tools. This design allows the backend to evolve independently, supporting future channels like mobile apps or kiosks.
The backend stores customer profiles, orders, menus, and locations in DynamoDB, and uses Amazon Location Service for geocoding and routing. The system recognizes callers by hashing their phone numbers, enabling personalized greetings without storing PII in plain text. The deployment process involves AWS CDK scripts that set up all components, including SIP gateways, agent containers, and telephony infrastructure.
This implementation demonstrates a scalable, flexible, and customer-friendly approach to automating restaurant orders, reducing staff workload, and improving customer experience during busy hours. It also illustrates how AWS services can be integrated to build complex, real-time voice applications with minimal latency and high reliability.
This solution showcases a scalable approach for enterprises to implement AI-powered telephony systems, potentially influencing customer service automation in the hospitality industry.