Natera’s intelligent appointment scheduling with Amazon Bedrock AgentCore
aws ·
Natera, a diagnostics company specializing in cell-free DNA testing, developed an automated voice agent using Amazon Bedrock AgentCore to streamline patient appointment scheduling for its mobile phlebotomy services. The system replaces manual scheduling calls and uses a dual-WebSocket bridge to separate telephony and model inference, along with event-driven latency masking that generates contextual filler responses during tool calls. It also employs a progressive trust model that gradually authenticates patients during the conversation. Natera migrated the workload from Amazon ECS to the AgentCore runtime, overcoming challenges related to WebSocket lifecycle and session state continuity by implementing connection pooling and externalized state management. During validation across 500 simulated calls, the agent achieved 100% tool-calling accuracy, sub-7-second perceived latency, and a cost of less than $0.01 per completed call. The architecture also provides built-in observability, enabling detailed tracing of each step for performance monitoring and issue resolution. The post explains the rationale behind these design choices and how the system maintains compliance and accuracy in a healthcare context.
aws