Building the thing nobody asked for

By 2025 I'd been at the company two years, and that's the year the model pages were handed to me. An inherited codebase, a build process I was still learning, and I was slow at it. Not all of the 2025 model pages got finished, and the dealerships noticed. "Where's our page? Why hasn't it been built?" That sting is where this project starts.

But before any pages could be built, someone had to know a car existed. Here's what that looked like. A new model just arrived at one of our seventeen rooftops. Find it.

The obvious fix existed. Dealership websites run on inventory feeds; the data was right there, powering the very pages I was checking by hand. So we asked if we could be added to one. The answer was no. We asked again later. No. Every route to the data came back closed, and the weekly chore stayed a chore.

Months later, mid-build with momentum running hot, I went at the wall directly: find the one person at the site vendor who could actually hand over feed access. Not one feed, it turned out. Seventeen, one per rooftop.

The chase worked. Seventeen feeds landed, which raised one last question: where do they live? That's the twist. The answer had been sitting in-house the whole time: a hosting server the company already ran, quietly capable of serving all seventeen feeds at stable URLs. I pointed my scan script at them and it came alive. The last problem was solved years before I asked the first question.

The idea sat in the back of my head for months before I wrote a line. What tipped it: I realized I'd already answered the hard questions without noticing. I'd shipped this very portfolio on GitHub and Netlify, so hosting a static app was a solved problem. The only open question was the code, and I answered that with AI as a pair programmer: I drove the product decisions, the interaction design, and the data model; it helped me move at the speed of my ideas instead of the speed of my syntax.

Rooftop+ opens to My Work. Not a dashboard, not a settings page: your queue. Every role gets their own: the builder sees pages ready to build, the SEO writer sees pages that need copy, the AEO writer sees the answer-engine layer. Tasks group by brand, clicking one opens a full-width focus mode with next steps and one-click actions, and owners assign themselves the moment someone touches a task. The design bet of the whole app: nobody should have to ask "what should I be doing?"

The second bet came straight from the 2025 sting. The reason missed pages hurt is that a car on the lot with no page is invisible money. So the shared board sorts by consequence, not by date.

And because a three-person team shipping hundreds of pages a year deserves to see it: Wins. Counters, a brand trophy shelf, a leaderboard computed from real task ownership. The chore got automated; the pride got a page.

I gave the first build to Bob, my manager, and James, our SEO writer. They loved it. Bob's words: "This is a great idea!" I appreciated that, but I'd known it was a good idea for months, for one specific reason:

Three months in, still fine-tuning. The pages this pipeline feeds have a system of their own: that's the vehicle page system case study, and it's the other half of this story: 2025 taught me to build pages faster; this app made sure I always know which page to build next.

About this demo: the embedded app is the real product with one swap. "Redwood Motors Group," its rooftops, team, and data are fictional, and names here are changed. The production build runs privately with live feeds, Google sign-in, and real-time sync.