Aurora
Aurora is the operating layer for a global logistics company moving more than four billion units annually. We replaced a constellation of legacy schedulers, custom EDI adapters, and human-in-the-loop spreadsheets with a single composable orchestration system.
01 — Problem
The client operated across forty-eight countries on a system that had grown by accretion for nearly twenty years. No single team could describe end-to-end flow. Every regional change required eleven coordinating tickets and a window of tolerated downtime.
The cost was not the software. The cost was the inability to think clearly about the business while the software dictated its own terms.
02 — Approach
We began by mapping the operational reality before touching code. Eight weeks of structured observation produced a domain map with clearly named seams — places where ownership and contracts had quietly diverged.
From there we designed Aurora as a control plane over a small number of well-defined services, with the legacy systems retained as adapters until each could be retired without ceremony.
03 — System
Aurora is a typed, event-sourced orchestration runtime with a declarative routing layer. AI-assisted exception handling reduces the volume of human intervention by an order of magnitude without removing human judgement from consequential decisions.
Every state transition is observable, every decision traceable, every contract versioned. The system is designed to be inspected, not just operated.
04 — Outcome
Eighteen months after first deployment, Aurora moved more than 92% of flows that had previously required manual coordination. Mean time to introduce a new lane fell from six weeks to thirty hours.
The work continues — but the operating company can, for the first time, describe its own system in a single sentence.
Measured, not asserted.
Manual flow elimination
92%
New-lane provisioning
6w → 30h
Production incidents (Y/Y)
−71%