Employee
Vacation
Tracker
An experiment in AI-assisted full-stack development. Built with Google AI Studio, integrated with Supabase Auth + DB, deployed via Claude Code and Netlify. Currently on hold — not a finished product.
The Starting Point
At a previous job, vacation requests still ran through email and Excel.
Employees would email a request, an admin would manually update a spreadsheet, then reply with approval or rejection. No visibility for employees on their remaining balance. No easy way for managers to spot conflicts. Just a slow, error-prone loop — for something that happens every week.
This became a test case for exploring AI-assisted development end-to-end: from UI scaffolding in Google AI Studio, to real backend integration with Supabase, to deployment via Claude Code and Netlify.
Core Features
Employee View
Self-Service Tracking
Employees see their total allocation, days used, remaining balance, and submit vacation requests — all without emailing HR.
Admin View
Approve or Reject
Managers see all pending requests and approve or reject with one click. No spreadsheet, no email thread.
Team Management
Add & Remove Staff
Admins can onboard new employees and remove departing ones — keeping headcount and leave entitlements in sync.
Auth & DB
Supabase Integration
Real authentication, persistent data, and multi-user sessions — connected and working. Replaced full localStorage mock.
Status
- UI built with Google AI Studio — React + TypeScript app scaffolded via vibe coding
- Core features working — employee tracking, admin approval, add/remove staff
- Deployed live — hosted on Netlify via Claude Code + GitHub
- Supabase integration complete — real auth (JWT), persistent DB, role-based access
- Known issue — re-adding a deleted employee with the same email fails (auth record persists)
- On hold — workflow needs significant rework before production use
What I Learned
Watching AI build something you don't fully understand is uncomfortable — and that's a useful signal.
This project was a deliberate experiment: how far can AI-assisted development go without deep foundational knowledge? The answer is surprisingly far — but the experience of watching code get written and deployed without fully understanding each decision felt off. Efficient, yes. But not the right kind of learning.
The real takeaway isn't "AI can't build things." It can. It's that relying on automation without the underlying knowledge to verify, debug, and own what gets built is a fragile position. This project will stay on hold until there's a clearer understanding of the stack it runs on.
Tools