Inspiration

What it does

Inspiration Missed appointments and administrative friction cost the healthcare industry over $150 billion annually in the US alone. Patients often dread navigating complex patient portals or waiting on hold to reschedule a visit, while clinic staff spend hours manually calling and texting patients.

We asked ourselves: What if healthcare communication felt as natural, immediate, and accessible as sending a WhatsApp message or SMS, powered by an AI assistant that actually understands clinic schedules and coordinates care end-to-end?

That inspiration drove us to build CareConnect AI for the AWS CDS Agentic AI Partner Hackathon—uniting Amazon Bedrock with AWS Communication Developer Services to make clinical appointment management effortless for patients and providers alike.

What it does CareConnect AI is an intelligent, multi-channel healthcare appointment and follow-up assistant.

Conversational Rescheduling & Booking: Patients can text "I can't make it tomorrow, can I reschedule for Friday morning?" via WhatsApp or SMS. The agent checks real physician calendars, proposes open slots, and confirms changes in natural language. Strictly Grounded Tool Execution: The AI never hallucinated slots or invents doctor availability; it invokes 7 strictly validated backend tools (getAvailableSlots, bookAppointment, rescheduleAppointment, cancelAppointment, etc.). Cross-Channel Continuity: A reminder is sent via WhatsApp, the patient reschedules in chat, an instant WhatsApp confirmation is issued, and an automated Amazon SES email arrives with clinical preparation instructions (arrival times, ID/insurance requirements, fasting instructions). Proactive Reminders & Consent Enforcement: Background workers dispatch timely reminders only to patients with active opt-in consent, respecting their preferred communication channel. Provider SaaS Dashboard: Healthcare staff get a clean, minimal dashboard to manage appointments, inspect doctor availability, view live conversational transcripts, and track delivery receipts across WhatsApp, SMS, and SES. How we built it AI & Agentic Reasoning: Built on Amazon Bedrock using the Bedrock Converse API and tool configurations (with Claude 3.5 Sonnet). Tools are bounded by Zod schemas, ensuring parameters are sanitized before hitting database functions. Communication Layer (AWS CDS): AWS End User Messaging Social (@aws-sdk/client-socialmessaging) for two-way WhatsApp Business messaging and webhook processing. AWS End User Messaging SMS (@aws-sdk/client-pinpoint-sms-voice-v2) for transactional SMS alerts. Amazon SES (@aws-sdk/client-ses) for responsive, branded HTML clinical preparation emails. Backend Architecture: Node.js, Express, and TypeScript connected to Prisma ORM with relational models for Patients, Doctors, Availabilities, Appointments, Conversations, Notifications, and Audit Logs. Frontend Dashboard: React 18, Vite 5, Tailwind CSS, and Lucide icons—crafted with a restrained, clinical palette (slate, sky, and teal) designed for high usability and accessibility. Reliability & Dev Sandbox: Implemented real AWS SDK v3 calls alongside an intelligent local Development Simulation Mode, allowing complete zero-cost sandbox testing. Challenges we ran into Preventing AI Hallucinations in Clinical Schedules: LLMs naturally tend to make up plausible-sounding times. We solved this by strictly decoupling reasoning from scheduling: the Bedrock agent is forced to call getAvailableSlots before suggesting any appointment, and database-level constraints prevent any double booking. Conversation Context Across Channels: Ensuring the AI recognizes patient context (phone number, existing bookings, and consent status) without exposing Protected Health Information (PHI) required building a robust session resolution layer and identity verification checks. Guardrails for Sensitive Actions: We implemented an explicit confirmation flow for cancellations, requiring the patient to affirm their intent before the cancelAppointment tool can execute. Accomplishments that we're proud of True Multi-Channel Synchrony: Successfully completing the full round-trip workflow from an automated WhatsApp reminder to an AI-powered reschedule, live database update, and instant Amazon SES clinical preparation email. Zero Double Bookings: Robust calendar validation that factors in physician working hours, existing appointments, and appointment buffer times. Production-Grade Architecture: Modular service separation using AWS SDK v3, immutable audit logging, rate limiting, and complete schema-backed validation. Design Excellence: Building a polished, distraction-free SaaS dashboard that feels modern, accessible, and ready for actual clinic use. What we learned The Power of Bedrock Tool Calling: Amazon Bedrock's Converse API with structured toolConfig provides deterministic, reliable agent execution that makes agentic AI viable for high-stakes industries like healthcare. AWS Communication Developer Services (CDS): Unifying WhatsApp, SMS, and Email under AWS native services significantly reduces operational overhead and delivers real-time delivery transparency through event webhooks. Consent-First Design: In healthcare communication, respecting patient channel preferences and opt-in/opt-out status isn't just a legal compliance requirement (TCPA/HIPAA)—it's foundational to patient trust. What's next for CareConnect EHR/EMR Integration: Direct HL7/FHIR connectors to integrate with Epic, Cerner, and Athenahealth for live clinical schedule syncing. Multilingual Support: Expanding prompt engineering and Bedrock localized inference to support Spanish, Mandarin, and other languages seamlessly across WhatsApp and SMS. Clinical Triage Routing: Integrating validated medical triage protocols that safely escalate acute symptoms directly to emergency services or an on-call nurse triage line. Rich Interactive Messaging: Leveraging WhatsApp interactive button messages and list pickers for 1-tap slot selection.