AI DevOps & Infrastructure
Sonia Health: Agentic architecture design that scales
"As Sonia Health scaled our voice-based AI therapy platform, we needed to decouple our agent infrastructure from our core backend to support independent scaling and faster iteration. We partnered with Tech 42 to build that foundation on AWS.
The Tech 42 team designed and delivered a dedicated agent serving environment — fully provisioned as infrastructure-as-code using Terraform — with the networking, session management, and observability layers needed to support our real-time voice pipeline. The collaboration was smooth from start to finish.
The team was communicative, delivered a thorough knowledge transfer, and left us with solid documentation to manage and iterate on the infrastructure independently. We're heading into the next phase of this partnership with a clear architectural direction and a strong foundation to build on."
Jimmy Li
ML Engineer
Project summary
SoniaHealth delivers a voice-based AI therapy platform powered by real-time conversational agents. Their agent layer was coupled to core application infrastructure, blocking independent scaling and faster iteration on agentic workloads. Tech 42 built a dedicated agent serving environment on Amazon ECS Fargate, provisioned as infrastructure-as-code using Terraform. The architecture includes Route 53 DNS and an Application Load Balancer for ingress, Amazon ElastiCache Redis for session state and multi-agent coordination, DynamoDB for per-tenant configuration, private ECR repositories for container images, and CloudWatch for observability. Tech 42 also evaluated AWS Bedrock model alternatives against SoniaHealth's existing prompts and request/response pairs, measuring latency and quality.
