Natera, a global diagnostics company, has developed an intelligent appointment scheduling system using Amazon Bedrock AgentCore to improve the patient experience for oncology patients. The system allows patients to book phlebotomy appointments through a conversational voice agent, eliminating the need for manual scheduling calls. This approach maintains compliance and accuracy standards required in healthcare while offering a more convenient experience for patients. The solution is designed to address the challenges of scheduling at scale, including patient authentication, third-party vendor integration, and fallback options for complex scenarios. Source: awsml

The architecture of Natera’s voice scheduling agent is built on three core principles: a dual-WebSocket bridge pattern, an event-driven latency-masking technique, and a progressive trust model for mid-conversation authentication. The dual-WebSocket bridge separates telephony streaming from model inference, allowing teams to independently swap telephony providers or foundation models without redesigning the full system. The event-driven latency-masking technique generates contextual filler responses to maintain natural conversation flow during backend operations. The progressive trust model escalates authentication and memory access incrementally as the conversation progresses, creating a more natural interaction rather than a transactional experience. Source: awsml

Natera chose Amazon Bedrock AgentCore for its fully managed architecture, which alleviates the operational overhead of managing container infrastructure and scaling configuration. The service’s built-in observability allows teams to track individual sessions end-to-end, pinpoint latency issues, and quickly identify and fix problems without guessing. This visibility is critical in healthcare environments where accuracy and performance are non-negotiable. The migration from Amazon ECS to AgentCore runtime also preserved the design principles while transitioning to a serverless environment. Source: awsml