Roomy:TheBedspaceOS.

The Problem

Co-living in high-density cities like Dubai creates a logistical disaster at scale. I was living in a 21-person shared apartment — and the financial management was entirely manual. WhatsApp groups for rent reminders. Cash handoffs with no receipts. Laundry machines with no scheduling system. Disputes with no paper trail.

The core failure: no shared system of record for money, facilities, or accountability.

System Architecture

Before writing a single line of UI code, I designed the data architecture. The central challenge: building a relational data model without a backend, so the app would work instantly with zero latency.

The Pseudo-Relational Database Engine

I built a custom database layer on top of localStorage that mimics relational behaviour:

• Entities: Tenants, Rooms, Rent Ledger, Payments, Laundry Slots
• Relations: Foreign key-style references between tenant IDs and payment records
• Transactions: Atomic write operations that update multiple entities simultaneously (e.g., confirming a payment updates the ledger, the tenant balance, and the admin summary in one operation)
• Live Queries: A subscription model that re-renders UI components when underlying data changes — similar to Supabase's real-time, but fully client-side

This eliminated all network latency. Every interaction — checking a balance, confirming a payment, booking a laundry slot — responds in under 16ms.

Key Features

Smart Rent Ledger

The ledger auto-calculates each tenant's balance based on their room type, shared utility split, and any outstanding arrears from previous months. Admins see a summary view; tenants see only their own balance.

Role-Based UI

Two distinct interfaces share the same codebase:

• Admin view: Full ledger, payment confirmation, tenant management, arrears tracking
• Member view: Personal balance, payment history, laundry booking

Role is determined by a PIN-based login stored locally — no backend authentication required.

Live-Status Laundry Scheduler

The laundry scheduler enforces hard daily limits (max 2 bookings per tenant per day) and displays real-time slot availability. Conflicts are prevented at the data layer, not the UI layer.

What I Learned About Vibe Coding

This project was my deepest experiment with AI-assisted development using Cursor. The workflow that worked:

• AI for boilerplate — state management, CRUD operations, form handling
• Human for architecture — the database schema, the role system, the conflict resolution logic
• AI for debugging — identifying edge cases in the localStorage transaction model

The most important lesson: AI accelerates execution but cannot replace systems thinking. The pseudo-relational engine required understanding why relational databases work before I could replicate the behaviour without one.

This project was a deep dive into Vibe Coding. By using AI to handle boilerplate and complex state synchronisation, I was able to focus on the system architecture — and ship a zero-latency experience that feels like a native app.