Inspiration
Finding an appointment is only the start of a patient journey. An enquiry, staff approval and a confirmed calendar reservation are different events. Tandvia makes those differences clear while connecting patients with dental practices.
What it does
Tandvia is a working dental-appointment coordination prototype with a backend-backed Alexa+ experience simulation. All demonstration records are synthetic, not real patient data or live clinical bookings.
The simulated experience supports appointment discovery and handoff to a private Tandvia account. Patients submit an enquiry and clinic-configured intake answers with consent for the selected practice. Private details stay out of shared-speaker responses and notification text. Practice staff review the linked lead, and the demonstration calendar is checked before the patient sees a demo-only confirmation. An enquiry is never presented as a confirmed real-world appointment.
The demonstrated workflow includes immutable intake corrections, stale-update conflicts, booking confirmation, cancellation and notification withdrawal, with a server-enforced rescheduling cutoff. Optional staff assistance is disabled by default and limited to review-only suggestions; it cannot independently approve bookings or contact patients.
How we built it
React and TypeScript provide the web interface; Java/Spring implements workflow boundaries. PostgreSQL persists the synthetic workflow data, and Kafka is part of the backend architecture. Calendar transports have local-fixture tests but remain disconnected from real provider accounts. A shared React Native/TypeScript/Expo mobile codebase is in development; native-device builds and real push delivery are not part of the demonstrated acceptance.
This entry uses the permitted Alexa+ simulation path. It does not claim live Alexa integration, an actual MCP server or MCP compliance. No live model inference or AWS service integration is claimed.
Verification and challenges
The integrated source passed nine PostgreSQL regression tests and 24 HTTP access-control checks, plus the browser journey. Testing exposed and corrected stale-slot and JSONB form-history ordering defects. We verified linked leads, staff confirmation, immutable correction history, cancellation and notification withdrawal using browser and database evidence.
The isolated reviewer package independently passed a fresh dependency restore/build, local startup, five HTTP readiness checks, loopback-only bindings and normal shutdown. Its real expiry and partial-failure cleanup checks remain separate, not assumed from simulations. Source is pinned in a private repository shared by invitation with the official reviewers.
The narrated demonstration is 2 minutes 33 seconds, built from verified application scenes. Narration uses an AI-cloned voice with the speaker’s permission.
What we learned and what is next
Booking integrity needs authoritative availability checks, explicit state transitions and tests against the actual database, not only in-memory fixtures. Private-account handoff and human review are important boundaries for voice-assisted appointment discovery.
Production identity management, encryption, tenant isolation, live calendar/model integrations and mobile-device behaviour still require complete production validation. These prototype checks are not a security certification or approval for clinical use. Next steps are production security review, practice-authorized vendor integrations and device testing.
Tandvia is a separate entry and is not entering the AWS Builder or Open Source mini challenges. AWS Activate is not used for this demonstration.
Log in or sign up for Devpost to join the conversation.