Developed a Doctor Appointment Booking App for a Saudi Clinic Network to Reduce Missed Appointments by 35%

An integration adapter converts the native message formats of the Hospital Information System into a clean internal domain model that doesn't contain legacy field names. Patient, Practitioner, Schedule, Slot and Appointment data becomes HL7 FHIR R4 resources, and the HIS is the sole source of medical data.

• Doctor schedules stored as weekly availability rules plus exceptions for leave, clinic holidays, and prayer breaks
• Slots generated from those rules at search time instead of being stored in advance
• PostgreSQL exclusion constraint on doctor and time range, so the database rejects any overlapping appointment from any screen
• Redis slot holds with a time to live of a few minutes during checkout
• Optimistic locking with a version column on appointment rows
• Timestamps stored in UTC and displayed in Arabia Standard Time

• The app generates a UUID idempotency key per booking attempt
• A retry with a known key returns the stored booking, and the same key with a different payload is rejected as a conflict
• Confirmation waits for the payment provider's webhook and a valid signature
• Bookings stuck in pending enter a reconciliation queue with bounded polling, then release the slot if payment never completes
• The app shows "awaiting payment" and disables a second attempt until the state resolves

• Appointment states modeled as a finite state machine: held, pending payment, confirmed, rescheduled, checked in, completed, cancelled and no show, with only legal transitions accepted
• BullMQ jobs schedule reminders 24 hours and 2 hours before each visit
• Failed sends retry with backoff, then move to a dead letter queue with alerting
• Firebase Cloud Messaging and APNs for push, with SMS as the fallback when push fails
• Signed deep links in each reminder for one-tap confirm or cancel
• No-show risk score from booking lead time, earlier missed visits and reminder replies

• Cancellation publishes an event with a slot identifier and a version number
• Waitlist service selects the next patient whose branch, doctor and time window match
• Offers expire after a few minutes, then pass to the next patient in line

• Correlation identifiers propagated through gateway headers, service logs, queue job metadata and adapter calls
• OpenTelemetry tracing and structured JSON logs with masked personal fields
• Dashboards for slot search p50, p95 and p99 latency, booking success rate, reminder delivery rate, queue depth and no-show rate per branch and doctor
• Append only audit records for every booking change and staff action