AI Healthcare Contact Center

The AI Healthcare Contact Center handles inbound patient phone calls for healthcare organizations. A caller explains what they need in their own words, in English or Spanish. The assistant confirms the request, verifies the caller's identity (including caregivers authorized to act for a patient), answers questions from organization-approved information, and books, reschedules or cancels appointments directly in the scheduling system. Every call ends with a structured summary and an audit record.

It is designed for the people who run a healthcare call center. Agents get a browser-based workspace to answer, hold, mute and transfer calls, with a live transcript, masked patient context and after-call work. Supervisors see live queues, agent status, failed transfers, call search, recording review and reporting. Knowledge editors and administrators control approved scripts, routing rules, business hours, consent and recording policies, and identity-verification rules, with two-person approval for anything the phone line will say. For a clinic or health system, the goal is to absorb routine scheduling and information calls so staff can focus on the calls that need a person.

Knowing when to hand off is the core design principle. Clinical concerns, signs of distress, uncertain identity, complex billing, unsupported requests and anyone who simply asks for a person are transferred live to staff, with the conversation context attached so the caller doesn't have to repeat themselves. Urgent concerns are routed first. The assistant never gives medical advice and repeats only content that has been approved and versioned. Calls run through Amazon Connect and a Lex V2 bot with Lambda handlers; scheduling uses an Oracle Health FHIR R4 adapter with read-back confirmation and protection against double-booking; and the staff console is a Next.js app with live updates over a FastAPI back end.

Verification is extensive: a 44-case conversation evaluation in English and Spanish, 13 simulated end-to-end phone calls driven through the real Lambda handlers, browser acceptance tests with accessibility checks, load testing, and backup and restore. Scheduling reads have been verified against Oracle Health's public sandbox. The full AWS platform for staging and production is defined in CDK and passes its tests, and a public sandbox, where visitors can call a real phone number and sign in as any staff role, is ready to deploy with call limits and nightly resets. Nothing is in production yet and no real patient data is used; go-live is waiting on external inputs such as an AWS account, a business associate agreement and release authorization.