I build things to understand problems, then ship what beats the status quo.

The problem

The Singapore Nursing Board registers about 3,000 nurses and handles 5,000 applications a year with a team of 10. Getting registered took around six months, and licensure exam processing took four.

Most of those six months was waiting, not work. Nurses waited for a paper exam to be held in a city they could travel to. Then came about a month for the paperwork to go through, a wait for results, and another month of processing the registration once they passed.

What I did

I own PRS 2.0, the Professional Registration System, end to end, leading a 20-person cross-functional team with the Ministry of Health and the Board as regulators. Rather than digitise the existing flow, we sat with the Board to diagnose where the friction was and co-designed automated routing for applications.

The key decision

We automated the happy paths and deliberately left the unhappy ones with people. Where the rules were clear and could be applied deterministically, and we had structured data from authoritative sources such as Singpass, the Ministry of Home Affairs, local schools and our own historical records, the system decides. Anything that could lead to a rejection, or needs to be handled with care like removal from the register, still goes to a Board officer.

Outcome

Time to registration dropped from six months to one (83 percent faster) and licensure exam processing halved, from four months to two. We're now extending the platform to other professional boards, starting with the Singapore Medical Council, with a target of automating 30 roles by 2027.

The problem

A Figma prototype often took two to four weeks to build and take through users. Much of that time went into tedious work, like keeping copy consistent across screens.

What I did

I built a Claude Code agent that turns user stories into testable React interfaces. Instead of static mockups, users test something that behaves like the real product. Claude now handles the whole loop: building the prototype, taking notes during sessions, summarising findings and proposing UX fixes.

Because the prototypes are real code, we deploy them to an accessible website, generate every scenario we want to test, and let users work through them at their own pace.

Outcome

The UX team dropped Figma for prototyping. They can now produce a prototype within a week and finish user testing the week after. All of our user testing runs on prototypes from this agent, five rounds so far.

Testing real scenarios also changed the product. Users showed us that some flows were redundant, and the edge cases they hit led us to remove an entire flow and merge it into another.

The thesis

AI has collapsed the cost of building software, but vibe-coded apps fail in production on reliability, compliance and security. Small and medium enterprises can now build their own tools, but they can't afford to get those three things wrong.

What I'm building

Rekaflow is a no-code platform that lets small organisations stand up compliant record-and-process systems without a development team. I'm building it end to end with frontier models.

It's still in the build phase. I'll share more here as it takes shape.