The problem
Booking a cleaning, repair or appliance service usually means calling around or using a classifieds listing, with no guaranteed availability, no clear pricing and no accountability once someone turns up.
QuickServe models the alternative end to end: the customer picks a service with a price checked on the server, books a slot, pays online or cash on delivery, and the order goes to a pool of partners who can accept or decline. The customer then chooses among the partners who accepted.
The approach
- The backend uses clean, hexagonal architecture (Spring Boot 3, Java 21). Orders, payments, partners and users have no framework dependencies; MySQL, Redis, Razorpay, AWS SQS and Firebase each sit behind a port with an adapter.
- Dispatch is a real matching workflow on an explicit state machine, from pending through dispatching, partner assigned and confirmed, ongoing, to completed or cancelled. Payment status is tracked separately from order status.
- The catalogue is read from Redis, rebuilt on a miss and invalidated by events when a banner or price changes, so cached data doesn’t go stale silently.
- Razorpay payments are HMAC-verified twice: the checkout callback from the app and the asynchronous webhook, which is the source of truth.
- OTP login issues JWT access and refresh tokens typed at the claim level, so a refresh token can’t be replayed as an access token.
- The Flutter app runs on its own design system (colour, type, spacing, radius and shadow tokens) across all 14 screens, with a five-tab shell that keeps each tab’s state and scroll.
One fix worth keeping: newer releases of the Razorpay Flutter plugin broke the Android manifest merge and older ones used a removed API, so the plugin and its native checkout are pinned to a known-good pair, with the reason written down in the repo.
Where it is
A side project. The backend has been deployed, but there are no real users yet.