How Stretto built an AI voice agent with Amazon Nova 2 Sonic
Stretto created “Tara,” an AI voice agent that replaces web‑form bankruptcy intake with a phone‑based, speech‑to‑speech interview powered by Amazon Nova 2 Sonic, a foundation model available through Amazon Bedrock. By using a single model for both speech recognition and generation, the system reduces latency and delivers more natural, emotionally aware conversations while keeping data within AWS security controls. The agent extracts structured financial details in real time and applies attorney‑defined rules to qualify leads, enabling cost‑effective automation of a high‑value intake process.
⚡ Key Takeaways
- Amazon Nova 2 Sonic is a speech‑to‑speech foundation model accessed via Amazon Bedrock.
- The architecture eliminates separate speech‑to‑text, LLM, and text‑to‑speech components, lowering end‑to‑end latency.
- Sensitive financial data stays on the AWS network using VPC endpoints (AWS PrivateLink) and IAM‑based access control.
- Integration also includes Anthropic’s Claude through the same Bedrock managed service for any supplemental language tasks.
- The solution requires no on‑prem model infrastructure, relying entirely on Bedrock’s managed offering.
Engineers can now build low‑latency, privacy‑focused voice agents for regulated domains without managing complex multi‑model pipelines, accelerating time‑to‑value while preserving compliance.
✅ Practical Steps
- Provision Amazon Bedrock access and enable the Nova 2 Sonic model via a VPC endpoint (AWS PrivateLink) in your AWS account.
- Configure IAM policies to grant your application permission to invoke Nova 2 Sonic (and Claude if needed) within the VPC.
- Replace any existing speech‑to‑text → LLM → text‑to‑speech workflow with a single call to the Nova 2 Sonic API for end‑to‑end speech processing.
Want the full story? Read the original article.
Read on AWS ML Blog ↗