By April 2025, the platform had been in my head for the better part of a decade. The shape of it. The way it would feel to use. The exact place in the workflow where every existing healthcare staffing tool was failing the people I'd spent ten years working with at Actriv. None of that was the hard part. The hard part was that I had never had the right tools to build it without an engineering team I couldn't afford and timelines I couldn't justify.
For most of that decade, I watched DONs drown in chaos that no existing tool could touch — and I watched my own ideas for fixing it die at the door of every engineering estimate I couldn't afford. I had a finished platform in my head and no way to ship it. By April 2025, the carrying cost of that gap had become unbearable.
That month, three of us sat down to try anyway. Me, Oleksandra "Sasha" Matkovska running product, Dmitri Batulin running technical execution, and Kole Nelson on the founding team carrying the AI systems work that would become some of the platform's most important infrastructure in the months that followed. Lovable open in one window. A whiteboard full of workflows. Five days on the clock.
We built it.
The decision
Most of the decisions about ShiftNex had already been made by the time we started building. Ten years of running Actriv had given me a reasonably complete picture of what good would look like. Two patents — US 11,947,875 and the continuation that became US 12,327,067 — had codified the technical architecture for voice-AI scheduling and event-conflict resolution. The vision was settled.
What I didn't have, until April 2025, was the ability to build it without engineers.
Lovable closed that gap. We participated in a Lovable Ship competition during the sprint period — a constraint that turned into a forcing function. The deadline was real. The team was small. The features had to be ruthlessly prioritized. By the end of the week, the smallest possible version of the platform was live: a facility could post a shift, a clinician could see it and accept it, and both sides had a confirmation they could trust.
That was the foundation. Everything else was iteration.
Month one
The first thirty days after launch were where the real architecture got built. The five-day sprint produced a working prototype. The next month produced something approaching a real product.
We onboarded the first facility. Then the second. Then the tenth. Each new client surfaced a different exception we hadn't planned for — credentialing edge cases, EMR-specific compliance requirements, scheduling rhythms unique to particular clinical settings. Every exception became a pattern we taught the system to handle. Every pattern became a small piece of the platform's intelligence.
The AI agents that would eventually run on top of the platform — the ones built on Claude Code and orchestrated through Cowork — started taking on operational load by the end of month two. Credentialing checks. Shift confirmations. Compliance audits. Customer follow-up. The kind of work a traditional staffing company would have hired thirty people to do, running twenty-four hours a day without interruption.
The five-month mark
By August 2025, five months after launch, the platform had crossed thresholds I'm still surprised by when I think about them.
Roughly five thousand healthcare professionals were active on the platform. Annual recurring revenue had crossed one million U.S. dollars — reported in the Kenyan press as KSh 129 million, a number that traveled across diaspora media throughout late April 2026 as the story of how a Kenyan-American operator had used AI tooling to build a national U.S. healthcare staffing platform without an engineering team.
The press attention was a strange thing to watch from the inside. Most of what I read about myself in those months was being written by people who had never spoken to me — assembled from public records, LinkedIn profiles, and the press release that announced the platform. The story they told was approximately right but missing the layer that mattered most: the five months weren't really about the five months. They were about the ten years before them.
Everything we built that summer was built fast because the actual hard problems had already been solved — slowly, painfully, expensively, over a decade of running a staffing company. Every workflow we coded, every match algorithm we tuned, every edge case we handled was the result of ten thousand hours of staring at the problem from inside the industry. The five months were just the moment when AI tooling caught up to the operating context I'd been carrying around for years.
What the numbers meant
I want to be careful with how I talk about the numbers, because the temptation when you have early traction is to make the numbers the story.
The numbers were not the story. The story was that an operator with a decade of context, working with two collaborators and one team member who would later move on, used a no-code AI building tool to ship a platform that immediately served thousands of healthcare workers — workers who were already underserved by every existing staffing technology in the market. The fact that the platform crossed a million in ARR in five months wasn't impressive because of the dollar amount. It was impressive because it demonstrated that AI-native solo and small-team founders can now build inside regulated industries that used to require ten million in venture capital and a forty-person engineering team to enter.
That's the thesis. The numbers were just the proof.
What came next
The story since August 2025 is the story of compounding. The platform kept growing. The team stayed small. The patents that had been filed in 2023 became granted patents in 2024 and 2025. The agent layer running on top of the core platform took on more and more of the operational work that would have required dozens of human coordinators a generation ago.
The current state of the platform is on the homepage, and on the rest of this site. The numbers there are different from the numbers on this page — they should be, because time has passed and the platform has continued to grow. This page is a primary record of what was true between April and August 2025. The press at the time told the story in fragments. This is the version with the operator's perspective intact.
If you came to this page from a press article, the numbers above are the same numbers you read there. The reporting was accurate. What you didn't get from the press was the texture — the decade of context, the team that built it, the constraints that turned into forcing functions, and the operating thesis that made any of it inevitable in retrospect.
That texture is what this page is for.
