Seven years on call for one platform

US trucking companies are legally required to run DOT drug-and-alcohol testing programs: pre-employment tests, random pools, post-accident protocols, FMCSA Clearinghouse queries, motor-vehicle-record checks. The platform turns that regulatory maze into software. Fleet companies enroll drivers, order tests, collect e-signatures, and stay auditable.

The client hired me off Upwork in April 2019 for what could have been a small job. It became the longest engagement of my career. I architected the platform, built it, deployed it. I have been its entire engineering department ever since. For seven years, I have handled features, fixes, and late-night server calls. Same client, same codebase, same person answering.

• Regulated recordkeeping. Test results, consent forms, and status changes are compliance artifacts. Losing history isn't a bug; it's a liability for the customer.
• Five kinds of users with real power differences. A super admin and a company team member must never share an inch of accidental surface area.
• Agents work across companies. Some users legitimately operate many fleets. Scoping mistakes here would leak one company's drivers into another's dashboard.
• Sole maintainer. Every clever trick I add, I maintain alone, potentially years later. Cleverness has a carrying cost.

The platform charges companies in three ways: recurring subscriptions for platform access, per-order charges for tests and records as they're consumed, and one-time fees like driver enrollment. Setup intents handle card capture; webhooks reconcile all three streams back into a coherent billing state, including the awkward cases where a subscription lapses mid-order.

I built each charge type where it was needed, and the logic spread across controllers. It works, verified by seven years of production billing. A unified billing abstraction from the start would have made every Stripe feature since cheaper. That abstraction is on the migration roadmap.

The platform started as server-rendered Blade with Vue 2 sprinkled in. That was the right choice in 2019. It is becoming an Inertia and Vue 3 SPA now. The migration is happening one page at a time, with both stacks running in production and sharing auth, sessions, and layout.

Doing this live taught me more about migration strategy than any greenfield project could. Keep the boundary explicit. Migrate whole user journeys rather than random pages. Resist "while I'm here" rewrites. The migration is the work; improvements ride the next train.

3,406 fleet companies. 8,859 actively managed employees. FMCSA Clearinghouse queries, MVR consent flows with e-signatures, state-specific pricing, a built-in support ticket system, shipped and maintained by one person without a major incident across seven years.

This kind of platform does not need a dramatic launch story. The useful measure is simpler: after seven years, the client still sends the next feature request to the same developer because the last three hundred went fine.

• Solo longevity is an architecture requirement. Every decision gets judged by "will I understand this in year five?"
• State machines should be explicit from the first record. Retrofitting history is an order of magnitude harder than recording it.
• Scattered billing logic is technical debt with compound interest. Unify the abstraction before the third charge type, not after.
• Long-running systems deserve deliberate documentation. Seven years of context in one head is a bus-factor problem I actively write my way out of.