It was a Thursday afternoon.

Thousands of riders across Lagos were moving across the city, picking up packages, completing deliveries, and accepting new trips. Relay Pilot had finally reached the stage every product team hopes for: real usage at real scale.

Then I watched the production database memory graph climb.

45%.

60%.

85%.

100%.

MongoDB started throwing WiredTiger errors. The kind of errors that politely tell you to "retry the operation" while quietly warning that something is very wrong.

At that moment, customers were waiting for deliveries, riders were updating trip statuses, and logistics companies were monitoring operations from their dashboards.

A single bad query under enough load can bring all of that down.

Fortunately, it didn't.

Within minutes we identified the issue, deployed a fix, and stabilized the platform before the next traffic spike arrived.

That incident reminded me of something I've learned repeatedly while building infrastructure:

Production systems don't survive because of heroics. They survive because of engineering decisions made months earlier.

This is the story of how I built Relay Pilot.

The Problem We Were Solving

Logistics in Africa is fragmented.

Restaurants build their own dispatch systems.