All work

Independent products

ClinicDesk AI

A clinic front desk, one conversation away

Live portfolio demo · fictional clinic
My role

Full-stack product engineering

Timeline

2026

Platform

Web / Desktop

Technologies
ReactTypeScriptNode.jsSupabasepgvectorOpenAIResend
CareWell’s live public website: clinic information, services and the entry point to reception chat
Actual public demo · CareWell is a fictional clinic. Open full screenshot ↗

01 / Overview

ClinicDesk AI combines a patient-facing clinic website, a conversational reception assistant and a private staff dashboard. Built around the fictional CareWell Family Clinic, it brings clinic-grounded answers, appointment scheduling and enquiry management into one product.

02 / The challenge

A clinic front desk needs to answer repeated questions and coordinate appointments without losing context. An AI assistant also needs clear limits: answers grounded in clinic information, explicit booking confirmation and data access scoped to the right clinic.

03 / What I built

  • Built a React patient experience and staff dashboard backed by a typed Express API.
  • Used clinic-scoped retrieval with Supabase PostgreSQL and pgvector, plus a bounded AI tool loop for knowledge search, availability and staff handoff.
  • Required review and explicit confirmation before booking, with atomic slot allocation to prevent double-booking.
  • Connected appointment notifications to a transactional email outbox with retry and reminder processing.
  • Applied verified staff authentication, clinic membership checks and row-level security; kept the no-account demo mode local-only.

04 / The outcome

A deployed portfolio demo showing the patient-to-staff workflow, with a documented architecture and automated tests for booking, access control and retrieval. The public website can be explored at the live project link.

05 / Context & limits

CareWell is fictional: this is a portfolio demonstration, not a medical service. The assistant is for clinic information and scheduling, not medical advice. Hosted AI responses, email delivery and production performance depend on the deployed services; no patient adoption or live-service benchmark is claimed.

Next project

AspireLearn

03 / A GOOD PLACE TO START

Open to a conversation

Have a problem
worth solving?