Urgent.News

What's breaking now, across thousands of outlets.

AI

How Stretto built an AI voice agent with Amazon Nova 2 Sonic

Stretto replaced its web-form lead capture with Tara, an AI voice agent built on Amazon Nova 2 Sonic that conducts bankruptcy intake interviews by phone and qualifies leads. See how Stretto, the AWS Generative AI Innovation Center, and CI&T designed the voice architecture, validated it for sensitive conversations, and moved it to production.

Stretto, a company that helps consumers considering bankruptcy, sought to build an AI-powered voice agent to streamline the intake process for potential clients. Traditional web forms for capturing leads were losing valuable detailed information, as visitors often abandoned the process before completing the form. Phone interviews with intake agents were deemed too costly due to staffing and training requirements.

The solution Stretto sought was an AI voice agent that could conduct the entire intake interview over the phone, collecting financial data in real-time and sorting leads based on attorney-defined rules. To achieve this, the team partnered with AWS and CI&T, an AWS Partner, to build an AI voice agent called Tara, powered by Amazon Nova 2 Sonic, a speech-to-speech model in Amazon Bedrock.

Tara's architecture was designed to reduce latency and provide more natural-sounding conversations. Instead of chaining separate speech-to-text, language model, and text-to-speech components, Amazon Nova 2 Sonic processed speech input and generated speech output natively in a single model. This architecture was critical for the sensitive nature of bankruptcy intake calls, as it maintained a seamless conversation flow.

The call flow for Tara started with Twilio, which opened a WebSocket connection into AWS. From there, traffic passed through AWS WAF and an Application Load Balancer to the voice bridge on Amazon Elastic Container Service (Amazon ECS) running on AWS Fargate, where Amazon Nova 2 Sonic was accessed through Amazon Bedrock. After the call, the transcript was handed off to an asynchronous processor using Anthropic’s Claude 3 Haiku on Amazon Bedrock to extract structured fields, which were then persisted in Amazon Aurora Serverless v2.

Qualified leads were then routed to attorneys based on jurisdiction. Supporting services included NAT gateways, VPC endpoints (AWS PrivateLink), Amazon ECR, AWS Secrets Manager, and Amazon CloudWatch. By utilizing Twilio for telephony and Amazon Nova 2 Sonic for speech processing, Stretto was able to achieve a 95% reduction in per-minute voice cost compared to its previous solution, while also reducing latency and improving interruption handling.

Written by urgent.news from AWS Machine Learning's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at aws.amazon.com →

More in AI

From reporting to reasoning: Getting enterprise data ready for AI agents

A recap of the inaugural AWS Quick User Group DC meetup, where data leaders explored getting enterprise data ready for AI agents.

  • AWS Quick User Group DC held in Arlington, Virginia on October 7, 2026
  • FIG modernized legacy architectures with governed data layer for AI insights
  • Hands-on exercises demonstrated rapid deployment of custom AWS Quick applications

More from Wednesday 18 November →