Amazon Nova 2 Sonic: как создать голосового агента для записи пациентов

Amazon released a detailed guide on creating a voice agent for medical institutions based on Nova 2 Sonic and Bedrock AgentCore. The agent independently calls patients, confirms appointments, and reduces no-show rates without operator involvement.

What the voice agent can do

The agent takes on routine calls with reminders about upcoming appointments. It conducts full-fledged voice dialogue: authenticates the patient, clarifies appointment details, and records the response in the scheduling system. The entire conversation cycle runs autonomously — registration staff only engage if the situation goes beyond standard scenarios.

Key agent capabilities:

• Patient voice authentication before the conversation begins
• Confirming, rescheduling, or canceling an appointment
• Collecting health information before the visit
• Escalation to a live operator for non-standard requests
• Parallel processing of hundreds of calls at scale

This architecture allows clinics to handle all reminders without increasing staff and at any time convenient for patients — whether in the evening or on weekends.

How it's built

At the core of the stack is Amazon Nova 2 Sonic, a model with native support for two-way voice in real-time. Unlike classical systems where text is first recognized, then processed, and then synthesized back, Nova 2 Sonic works directly with the audio stream — this reduces latency and makes the dialogue more natural.

Orchestration is handled by Bedrock AgentCore: it manages calls to external functions — schedule verification, recording changes, retrieving patient history — and directs the next step of the conversation based on the data received. The guide includes a browser interface for development and testing: developers can walk through the entire scenario without actual telephony. For production, AWS recommends integration with Amazon Connect, but in the tutorial this layer is set aside — the focus is specifically on agent logic and voice interaction.

Why this matters for medicine

"You process routine calls at scale — this helps reduce the percentage of missed appointments," the official solution description states.

No-shows are a chronic healthcare problem: by various estimates, 15–30% of bookings are canceled without notice. Clinics lose revenue, and patients with genuine need remain without a slot. Automated voice reminders with the ability to immediately reschedule an appointment during the call reduce this metric without additional burden on staff. An additional benefit is collecting health information before the visit. If a patient reports a deterioration in condition, the doctor gets context in advance: it's possible to adjust the appointment protocol or offer an alternative format — telemedicine or an earlier slot.

What this means

A ready-made template from AWS significantly lowers the entry barrier. Implementing a voice agent for reminders is now possible without an ML engineering team — it's enough to adapt the example to your scheduling system and telephone provider. Healthcare is becoming one of the first sectors where voice AI agents transition from pilot projects to standard operational practice.