Smart Retail

Inflyter was a digital duty-free platform where travellers ordered products online and collected them at the airport. We wanted to add autonomous collection: Amazon-style lockers where customers could pick up their orders using a QR code, no staff interaction required.

The locker hardware came from an external supplier who also provided a basic web application to manage the workflow, the only option at the time. I had already built the entire Inflyter platform single-handedly (backend, two native iOS apps (B2C and B2B), cloud infrastructure) and was still maintaining all of it while managing client and supplier relationships, so building custom locker software wasn't yet feasible.

The supplier's app proved the concept, but it was designed as a generic locker management tool, not a platform we could extend with custom features. The codebase was unstable, difficult to extend or debug, and the interface wasn't suited for client-facing use.

After deploying two lockers in Beirut, it was clear: the concept worked, but we needed custom software to scale it.

The supplier's system assigned compartments randomly: a small perfume could block a large compartment meant for bulkier items, wasting capacity.

I built a reservation engine that calculated order volume from known product dimensions and allocated the optimal compartment, or combination of compartments, to fit it. This was a bin-packing optimisation problem at its core: maximise locker capacity while guaranteeing every order fits. It measurably improved daily throughput per locker.

Each locker contained hardware from different suppliers, none of which spoke the same protocol: touchscreen, payment terminal, barcode scanner, passport scanner, receipt printer, router, PCBs controlling the compartment lock solenoids, infrared presence sensors, a security camera, and speakers.

I designed a hardware abstraction layer that sat between the application logic and the physical devices. The application worked with clean, uniform commands ("open compartment 4," "verify passport," "print receipt") while the abstraction layer handled the specifics of each device underneath. If we needed to swap a payment terminal or barcode scanner for a different model, only the adapter changed. The business logic never had to be touched.

When airport retail stores faced restricted hours and closures during COVID, we saw an opportunity: if the locker could handle autonomous collection, why not autonomous purchasing?

I extended the API and backend to handle stock management, inventory, and direct-sale transactions, and designed the changes to the Windows application to support product browsing and purchasing, none of which existed before. We deployed the world's first digital duty-free vending machine in Geneva, then Luxembourg. The vending machines quickly started generating more revenue than the Inflyter app itself.

Smart Retail was no longer a feature. It was a product.