API keys stored, with every model call authenticated through AWS IAM
AI products on one HIPAA-aligned AWS foundation
Session audio sent to servers, since speech-to-text runs in the browser
Attunica
Delaware, United States
Behavioral Health Technology / Clinical and Academic AI Tools
Amazon Bedrock (Claude), AWS Fargate, Amazon Aurora Serverless, Amazon RDS (PostgreSQL), Amazon S3, AWS Secrets Manager, Amazon CloudWatch, AWS IAM, AWS STS, AWS Key Management Service, Amazon ECR, AWS Certificate Manager, Application Load Balancer
Attunica operates two connected AI products for behavioral health: a clinical note-taker that helps therapists manage session documentation, and a simulation platform that trains Master of Social Work students through live interactions with AI-generated client personas. Avahi migrated both products’ AI calls onto Amazon Bedrock and hardened the infrastructure behind them, moving Attunica off direct API-key management entirely and onto AWS IAM-based authentication for every model call.
Attunica is a behavioral health technology company built in collaboration with a leading school of social work. Its platform addresses two connected problems in the field: therapists are overwhelmed by documentation requirements, and social work programs lack structured tools for training clinicians on AI-influenced client behavior.
Attunica’s two products, a clinical note-taker and a live AI-simulation training platform, both called their AI models directly through provider APIs, each holding and managing its own API keys. For products handling therapy session content and PHI-adjacent data, that meant credentials to manage and rotate, no consistent access control across either application’s model calls, and no unified path to the audit logging and encryption controls that clinical and academic use requires.
Attunica needed every AI call across both products to run through a single, auditable authentication path rather than scattered API keys. Amazon Bedrock let Avahi move both products onto Claude models using AWS IAM credentials exclusively, so no key is stored, rotated, or can leak outside AWS’s own access boundary. Each service runs on AWS Fargate behind an Application Load Balancer, with the simulator’s structured data on Amazon Aurora Serverless and the note-taker’s on Amazon RDS for PostgreSQL, both encrypted at rest with customer-managed keys and reachable only through session-logged, no-bastion access paths.
As a premier-tier AWS partner with hands-on experience migrating production AI applications onto Bedrock, Avahi could take on a cross-cutting migration that touched application code, infrastructure as code, CI/CD pipelines, and test suites across two independently maintained repositories, without treating either product as a green-field rebuild.
Avahi migrated both of Attunica’s products off direct model-provider APIs and onto Amazon Bedrock, authenticated entirely through the AWS credential chain rather than stored API keys.
For the clinical note-taker, both of its AI features, generating progress notes and treatment plans from clinician input, now call Claude through Bedrock from a single client used across the application. The application runs on AWS Fargate behind a load balancer, backed by an encrypted, single-instance PostgreSQL database on Amazon RDS with enforced TLS and deletion protection. Speech-to-text for clinician note-taking runs entirely client-side in the browser, so session audio itself never reaches Avahi’s or Attunica’s servers; only the resulting text is sent to Claude for note generation.
For the simulation platform, Avahi migrated three separate call sites, covering avatar-response generation, training-module content, and post-session evaluation scoring, onto Bedrock, while leaving the platform’s live voice and avatar layer on its existing real-time providers, which don’t route through Bedrock. The platform runs as multiple independent AWS Fargate services, each with its own access role, backed by Amazon Aurora Serverless for structured data. Avahi also built the evaluation pipeline that scores training sessions against a rubric, delivered and ready to activate once Attunica finalizes its own scoring policy. Across both products, Avahi applied HIPAA-aligned controls. Enforced database encryption in transit and at rest, full network traffic logging, session-logged database access with no standing shell access, and HTTPS-only traffic at every load balancer.
Both of Attunica’s products migrated from direct model-provider APIs to Amazon Bedrock, authenticated via AWS IAM with no stored API keys
A clinical note-taker generating progress notes and treatment plans via Claude, running on AWS Fargate with an encrypted PostgreSQL database
A migrated simulation platform running multiple independent services on AWS Fargate, with model calls migrated to Bedrock and structured data on Amazon Aurora Serverless
A rubric-based session evaluation pipeline, built and ready to activate pending Attunica’s scoring-policy decision
HIPAA-aligned controls applied consistently across both products: encryption at rest and in transit, full network traffic logging, and session-logged database access
Complete architecture and infrastructure documentation, plus knowledge transfer, covering both products
Attunica came out of this engagement with both of its AI products running on a consistent, auditable foundation instead of scattered, directly-managed API credentials. Every model call across both products now authenticates through AWS IAM rather than a stored key, and both products carry the same HIPAA-aligned controls for encryption, logging, and access. With full architecture documentation and knowledge transfer complete, Attunica’s team owns both systems going forward, including an evaluation capability for the training platform that’s ready to switch on as soon as the scoring policy behind it is finalized.
Attunica
Delaware, United States
Behavioral Health Technology / Clinical and Academic AI Tools
Amazon Bedrock (Claude), AWS Fargate, Amazon Aurora Serverless, Amazon RDS (PostgreSQL), Amazon S3, AWS Secrets Manager, Amazon CloudWatch, AWS IAM, AWS STS, AWS Key Management Service, Amazon ECR, AWS Certificate Manager, Application Load Balancer
Attunica operates two connected AI products for behavioral health: a clinical note-taker that helps therapists manage session documentation, and a simulation platform that trains Master of Social Work students through live interactions with AI-generated client personas. Avahi migrated both products’ AI calls onto Amazon Bedrock and hardened the infrastructure behind them, moving Attunica off direct API-key management entirely and onto AWS IAM-based authentication for every model call.
Attunica is a behavioral health technology company built in collaboration with a leading school of social work. Its platform addresses two connected problems in the field: therapists are overwhelmed by documentation requirements, and social work programs lack structured tools for training clinicians on AI-influenced client behavior.
Attunica’s two products, a clinical note-taker and a live AI-simulation training platform, both called their AI models directly through provider APIs, each holding and managing its own API keys. For products handling therapy session content and PHI-adjacent data, that meant credentials to manage and rotate, no consistent access control across either application’s model calls, and no unified path to the audit logging and encryption controls that clinical and academic use requires.
Attunica needed every AI call across both products to run through a single, auditable authentication path rather than scattered API keys. Amazon Bedrock let Avahi move both products onto Claude models using AWS IAM credentials exclusively, so no key is stored, rotated, or can leak outside AWS’s own access boundary. Each service runs on AWS Fargate behind an Application Load Balancer, with the simulator’s structured data on Amazon Aurora Serverless and the note-taker’s on Amazon RDS for PostgreSQL, both encrypted at rest with customer-managed keys and reachable only through session-logged, no-bastion access paths.
As a premier-tier AWS partner with hands-on experience migrating production AI applications onto Bedrock, Avahi could take on a cross-cutting migration that touched application code, infrastructure as code, CI/CD pipelines, and test suites across two independently maintained repositories, without treating either product as a green-field rebuild.
Avahi migrated both of Attunica’s products off direct model-provider APIs and onto Amazon Bedrock, authenticated entirely through the AWS credential chain rather than stored API keys.
For the clinical note-taker, both of its AI features, generating progress notes and treatment plans from clinician input, now call Claude through Bedrock from a single client used across the application. The application runs on AWS Fargate behind a load balancer, backed by an encrypted, single-instance PostgreSQL database on Amazon RDS with enforced TLS and deletion protection. Speech-to-text for clinician note-taking runs entirely client-side in the browser, so session audio itself never reaches Avahi’s or Attunica’s servers; only the resulting text is sent to Claude for note generation.
For the simulation platform, Avahi migrated three separate call sites, covering avatar-response generation, training-module content, and post-session evaluation scoring, onto Bedrock, while leaving the platform’s live voice and avatar layer on its existing real-time providers, which don’t route through Bedrock. The platform runs as multiple independent AWS Fargate services, each with its own access role, backed by Amazon Aurora Serverless for structured data. Avahi also built the evaluation pipeline that scores training sessions against a rubric, delivered and ready to activate once Attunica finalizes its own scoring policy. Across both products, Avahi applied HIPAA-aligned controls. Enforced database encryption in transit and at rest, full network traffic logging, session-logged database access with no standing shell access, and HTTPS-only traffic at every load balancer.
Both of Attunica’s products migrated from direct model-provider APIs to Amazon Bedrock, authenticated via AWS IAM with no stored API keys
A clinical note-taker generating progress notes and treatment plans via Claude, running on AWS Fargate with an encrypted PostgreSQL database
A migrated simulation platform running multiple independent services on AWS Fargate, with model calls migrated to Bedrock and structured data on Amazon Aurora Serverless
A rubric-based session evaluation pipeline, built and ready to activate pending Attunica’s scoring-policy decision
HIPAA-aligned controls applied consistently across both products: encryption at rest and in transit, full network traffic logging, and session-logged database access
Complete architecture and infrastructure documentation, plus knowledge transfer, covering both products
Attunica came out of this engagement with both of its AI products running on a consistent, auditable foundation instead of scattered, directly-managed API credentials. Every model call across both products now authenticates through AWS IAM rather than a stored key, and both products carry the same HIPAA-aligned controls for encryption, logging, and access. With full architecture documentation and knowledge transfer complete, Attunica’s team owns both systems going forward, including an evaluation capability for the training platform that’s ready to switch on as soon as the scoring policy behind it is finalized.
Let’s explore your high-impact AI opportunities together in a complimentary session
