A pet-services marketplace on serverless AWS — six categories, Stripe payments, and hosting costs that scale to zero.
Lead mobile & backend developer — freelance
PawsNeeds connects pet owners with verified providers across veterinary care, grooming, training, walking, boarding, and transport. Owners discover nearby providers on a map and book in-app; providers manage their catalogues; an admin verification workflow keeps the marketplace trustworthy.
The client was early-stage with spiky, unpredictable traffic — and a budget that couldn’t fund idle servers. I built the React Native app and the AWS backend, and made the architectural bet that defined the project.
Pay-per-request Lambda kept hosting near zero between bookings and absorbed evening spikes with no capacity planning. The cost: cold starts on the paths users feel — addressed by keeping hot-path functions small and never letting booking latency pay for the admin panel’s dependencies.
A single function is simpler to deploy, but couples scaling and bloats every cold start with every domain’s dependencies. Per-domain functions scale independently — search bursts don’t warm payment code — at the cost of more deployment wiring, which CI absorbed.
Optimizing before upload cut media bandwidth roughly 40% and removed an entire processing tier. The trade: trusting the client pipeline — mitigated with server-side validation of type and size.
Social auth or SMS OTP
“In my area” geosearch
Calendar + cart
Stripe, multiple methods
Ratings close the loop
Submit credentials
Admin approval workflow
Services, pricing, availability
Status + notifications
Serverless economics came with a UX tax: a Lambda that hasn’t run recently takes extra seconds to respond, and a booking flow that stalls at payment is a booking that doesn’t happen.
I attacked it from three sides. Functions were split per domain so hot paths stayed small — the booking function ships without the admin panel’s dependencies. Bundles were trimmed until cold init was dominated by the runtime, not my code. The app hides the remainder: optimistic UI on booking steps and prefetching the next step’s data while the user is still picking a time slot.
The result is a flow that completes in under two minutes on average, on infrastructure costing an estimated ~60% less than the equivalent always-on setup at the client’s traffic level.
YOU GOT SOMETHING IN MIND ?
Full Stack Developer
Open to roles in Europe & Tunisia — based in Tunis (UTC+1, Paris hours)