Healthcare provider Clarus Care is tackling patient communication bottlenecks by ditching rigid menu systems in favor of natural conversation. In collaboration with AWS, they built a sophisticated contact center prototype using Amazon Bedrock to orchestrate high-performance models like Claude 3.5 Sonnet and the Amazon Nova series.

Unlike standard phone systems, this architecture handles multiple patient goals simultaneously (multi-intent resolution), such as requesting a refill while asking about a bill. By leveraging a hierarchy of specialized AI agents, the system determines the urgency of a call immediately, ensuring critical medical concerns are flagged and routed to human staff without delay.

The backbone utilizes a "multi-model" approach to balance speed and intelligence. Claude 3.5 Sonnet handles complex reasoning, while Amazon Nova Pro and Nova Lite extract structured data and generate low-latency responses. This strategy keeps processing times under three seconds. The platform even resolves provider name variations, correctly matching "Dr. Smith" to "Dr. Jennifer Smith" through semantic understanding rather than strict database lookups.

Healthcare provider Clarus Care is tackling patient communication bottlenecks by ditching rigid menu systems in favor of natural conversation. In collaboration with AWS, they built a sophisticated contact center prototype using Amazon Bedrock to orchestrate high-performance models like Claude 3.5 Sonnet and the Amazon Nova series.

Unlike standard phone systems, this architecture handles multiple patient goals simultaneously (multi-intent resolution), such as requesting a refill while asking about a bill. By leveraging a hierarchy of specialized AI agents, the system determines the urgency of a call immediately, ensuring critical medical concerns are flagged and routed to human staff without delay.

The backbone utilizes a "multi-model" approach to balance speed and intelligence. Claude 3.5 Sonnet handles complex reasoning, while Amazon Nova Pro and Nova Lite extract structured data and generate low-latency responses. This strategy keeps processing times under three seconds. The platform even resolves provider name variations, correctly matching "Dr. Smith" to "Dr. Jennifer Smith" through semantic understanding rather than strict database lookups.