product updateAmazon Web Services

AWS releases healthcare appointment agent tutorial using Nova 2 Sonic speech-to-speech model

TL;DR

AWS published a technical guide for building voice appointment agents using Amazon Nova 2 Sonic, a speech-to-speech model that processes audio natively without separate transcription steps. The tutorial covers authentication, scheduling, and escalation tools running on Amazon Bedrock AgentCore with DynamoDB persistence.

3 min read
0

AWS releases healthcare appointment agent tutorial using Nova 2 Sonic speech-to-speech model

AWS published a technical guide for building voice appointment agents using Amazon Nova 2 Sonic, a speech-to-speech model available on Amazon Bedrock that processes audio natively without chaining separate transcription, reasoning, and text-to-speech services.

The tutorial addresses healthcare appointment no-shows, which range from 5–30 percent across US healthcare providers depending on specialty. The agent authenticates patients by voice, manages appointments (confirm, cancel, reschedule), collects pre-visit health information, and escalates to human staff when needed.

Technical architecture

The implementation uses Amazon Bedrock AgentCore, a serverless runtime for AI agents with IAM-authenticated WebSocket endpoints. The agent is built with the Strands Agents SDK's BidiAgent class for bidirectional voice streaming.

The system architecture includes:

  • React frontend capturing browser microphone audio over SigV4-signed WebSocket
  • Amazon Cognito for authentication with temporary AWS credentials
  • Amazon Bedrock AgentCore hosting the containerized Strands BidiAgent
  • Nova 2 Sonic processing speech input and generating audio responses
  • Three DynamoDB tables storing patient records, appointments, and time slots
  • Amazon SNS for escalation notifications

According to AWS, Nova 2 Sonic's native speech-to-speech processing retains vocal nuance like tone and hesitation that text transcription discards. The model supports 16kHz input and output sample rates with multilingual capabilities and accent handling.

Seven healthcare tools

The agent implements seven Python functions using the Strands SDK's @tool decorator:

  1. authenticate_patient: Verifies identity using name and last four SSN digits via DynamoDB Global Secondary Index, enforcing three-attempt limit
  2. confirm_appointment: Updates appointment status from Scheduled/Rescheduled to Confirmed with idempotent checks
  3. cancel_appointment: Changes status to Canceled with timestamped reason
  4. find_available_slots: Queries open time slots from provider schedule
  5. book_appointment_slot: Finalizes rescheduling after slot selection
  6. collect_health_info: Gathers pre-visit patient information
  7. escalate_to_human: Triggers Amazon SNS notification for human staff intervention

Nova 2 Sonic decides when to invoke each tool based on patient responses. The setup requires minimal code:

from strands.experimental.bidi import BidiAgent
from strands.experimental.bidi.models import BidiNovaSonicModel

model = BidiNovaSonicModel(
    region="us-east-1",
    model_id="amazon.nova-2-sonic-v1:0",
    provider_config={
        "audio": {
            "input_sample_rate": 16000,
            "output_sample_rate": 16000,
            "voice": "tiffany",
        }
    },
    tools=tools,
)

agent = BidiAgent(
    model=model,
    tools=tools,
    system_prompt=system_prompt,
)

Deployment details

The tutorial includes a browser-based interface for testing. To connect the agent to actual phone lines for outbound dialing, AWS notes that integration with a telephony service like Amazon Connect Customer is required.

The serverless deployment runs entirely on AWS infrastructure with Amazon Cognito authentication, DynamoDB persistence, and SNS notifications. The Strands SDK handles bidirectional streaming complexity, managing audio input, tool invocation, and audio response generation.

What this means

This tutorial demonstrates AWS's push to move speech AI workloads away from chained transcription-LLM-TTS architectures toward native speech-to-speech models. By preserving acoustic context through the full conversation, Nova 2 Sonic can potentially detect patient anxiety or confusion that text transcription loses.

The healthcare use case targets a specific economic problem: appointment no-shows cost providers revenue and delay patient care. Voice agents that handle routine confirmation calls at scale could reduce manual phone bank operations, though AWS provides no data on actual no-show rate reduction from this approach.

The modular tool design means developers can add capabilities by writing individual Python functions without rebuilding the agent. For organizations already using AWS infrastructure, the serverless deployment on Bedrock AgentCore eliminates separate infrastructure management.

Related Articles

product update

Google upgrades Gemini Live on Home speakers to speech-to-speech architecture, for Premium subscribers only

Google says Gemini Live on Google Home speakers and smart displays is moving to its "latest speech-to-speech architecture," promising "significantly faster response times." The upgrade is labeled "coming soon" and requires a Google Home Premium subscription, which starts at $10/month.

product update

Zed v1.30.0-pre adds GPT-6.1 Terra support and a commit message model picker

Zed's v1.30.0-pre release adds GPT-6.1 Terra to the OpenAI provider and a model picker for AI commit message generation in the Git Panel. It also adds an Extra High thinking effort level for Gemini 4 Hydrogen and reduces GPU usage across the app.

product update

Zed v1.29.1 adds Claude Fable 5.2 for Anthropic BYOK users and a Git Panel branch mode

Zed v1.29.1 adds support for Claude Fable 5.2 for Anthropic BYOK users, plus Claude Mythos 5.2 for organizations with access. The release also adds a Git Panel branch mode, a bookmark picker, configurable Git date formats and a setting to turn off automatic agent thread titles.

product update

Replit brings Microsoft Fabric, Teams and local work to general availability; Azure publishing in early access

At Microsoft Ignite, Replit announced general availability for its Microsoft Fabric integration, local access in Replit Desktop, and a Microsoft Teams integration. Publishing apps to a company's Azure tenant is in early access, with Full BYOC planned next. Replit was also named 2026 Microsoft Partner of the Year for Startups.

Comments

Loading...