7-Eleven · DEX/FuelControl · 2021—present
Forecourt payment & IoT platform
Secure fuel-station automation across major fuel retail brands: EMV payment at the dispenser, edge state machines, and an AWS backbone that keeps a national fleet connected.
Java / Spring Boot · AWS IoT Core · Kinesis · DynamoDB · Lambda · ECS · Node.js
Probe a stage — hover or Tab.
Problem
A fuel dispenser is a payment terminal bolted to a pump on the edge of a retail network. It has to accept EMV card payments, obey fuel-control commands, and report what it dispensed — over links that drop, at sites with decades-old wiring, across hardware from different manufacturers.
Legacy forecourt integrations treated each site as an island: polling-based status checks, manual reconciliation between what the pump dispensed and what the point of sale recorded, and truck-roll troubleshooting when a device went quiet.
Constraints
- EMV and PCI certification boundaries fix where card data may flow; the architecture has to work around them, never through them.
- Connectivity is intermittent by nature — a site must keep fueling and settle correctly when the link returns.
- Dispenser hardware is heterogeneous across brands and generations; the platform speaks to all of it through one contract.
- Fuel doesn't stop: rollouts happen against a live national fleet with zero appetite for downtime.
Architecture
At the edge, site controllers run Java/Spring state machines that model each dispenser transaction explicitly — every pump state, payment state, and failure path is a first-class transition, not an exception handler. Events are buffered store-and-forward, so a dropped link degrades to delayed telemetry rather than lost transactions.
Edge devices authenticate to AWS IoT Core over MQTT with per-device identity. Telemetry fans into Kinesis for stream processing; hot state lands in DynamoDB, operational history in MongoDB. Commands flow the other way with explicit acknowledgement and idempotency keys, because a command that might have been applied is worse than one that failed.
Configuration and firmware rollouts move through the same command channel in waves, with per-site health gates — a bad rollout stops itself before it becomes a national incident.
Outcome
- The platform runs fuel-station automation across major fuel retail brands in the United States.
- Reconciliation between dispensed fuel and recorded transactions became an automated stream instead of a manual back-office task.