How Stretto built an AI voice agent with Amazon Nova 2 Sonic
Machine Learning Blog
This article describes how Stretto built Tara, an AI voice agent powered by Amazon Nova 2 Sonic, to replace web forms for bankruptcy intake interviews and improve lead quality for attorneys.
- Amazon Nova 2 Sonic processes speech input and generates speech output natively, reducing latency and enabling natural conversations for sensitive financial discussions
- Tara conducts multi-phase intake interviews collecting personal information, financial details, assets, and employment history while adapting tone to caller emotional state
- Architecture uses Amazon ECS on Fargate for the voice bridge, Amazon Aurora for data persistence, and Claude 3 Haiku for post-call structured data extraction
- Team validated voice quality first, deployed minimal intake to production, then expanded to 18+ categories with conditional branching based on real-world data
- Achieved 96% data extraction accuracy, ~8 minute average call duration, 20% increase in lead conversion rate, and 95% reduction in per-minute voice costs
- Implemented comprehensive testing including happy path, adversarial, compliance, and barge-in scenario testing for sensitive financial conversations
Stretto's phased approach demonstrates how speech-to-speech AI can deliver business value in high-stakes intake processes while maintaining quality and compliance standards.
The AWS News Feed is currently looking for gold sponsors. If you want to support the AWS community and reach a large audience of AWS professionals, consider sponsoring the AWS News Feed.
Related articles
2025
2026
2026
2025
The AWS News Feed is currently looking for silver sponsors. If you want to support the AWS community and reach a large audience of AWS professionals, consider sponsoring the AWS News Feed.