Clinical Scheduling App — From Zero to Live PWA in 3 Weeks
3 Weeks
Time to Ship
2-3 Hours
Daily Time Saved
60+
Commits
Designed and built PrepSync — a mobile-first progressive web app that replaces 2-3 hours of daily administrative overhead for a solo gastroenterologist. Handles scheduling, patient prep PDF generation, WhatsApp sharing, Google Calendar sync, and financial tracking. Went from zero to deployed on Vercel in 3 weeks with 60+ commits, all built through AI-assisted development. The doctor uses it daily on his iPhone.
The Challenge
A solo gastroenterologist in Mérida was spending 2-3 hours every day on manual admin: scheduling procedures, generating patient prep instructions (fasting rules, medication suspensions, special diet requirements), sending PDFs via WhatsApp, managing his Google Calendar, and tracking procedure finances. No receptionist, no admin staff — just a doctor with an iPhone between back-to-back procedures.
He needed something that worked on his phone, understood his clinical workflow, generated branded PDFs matching his existing templates, and didn't require him to learn new software. The budget and timeline ruled out a traditional app development process.
What I Delivered
- Complete PWA deployed on Vercel, optimized for iPhone Safari
- Clinical logic engine handling modifier stacking: diabetes fasting calculations, anticoagulant suspension dates, elderly patient prep modifications
- Branded PDF generation matching the doctor's existing clinical templates
- Two-way Google Calendar sync with procedure type auto-detection from free-text event titles
- Financial tracking dashboard with confirmed vs. pending revenue splits and Apple Health-style chart visualizations
- WhatsApp sharing integration for patient prep PDFs
- Backup/restore system and onboarding wizard
- Spanish date formatting and bilingual interface
- 400-line product specification (CLAUDE.md) serving as machine-readable design system and clinical logic reference
The AI Workflow
Wrote a 400-line product specification (CLAUDE.md) covering design system rules, clinical logic, and architectural decisions — then used Claude Code to implement the full stack: Next.js App Router, Dexie.js local database, client-side PDF generation, Google OAuth, and the clinical logic engine. All 60+ commits over 3 weeks were built through AI-assisted coding.
The non-obvious AI usage: used Claude not just for code generation but for encoding complex clinical domain logic — translating fasting protocols, drug suspension calculations, and modifier stacking rules into software requirements. The CLAUDE.md specification served as a machine-readable product spec that let AI build correctly without constant hand-holding.