← All work

Werlabs

Two production services owned end to end at a regulated Swedish health-tech provider: the appointments and booking service behind its paid video consultations, and a TypeScript/Playwright PDF pipeline that replaced a fragile legacy JVM stack.

Werlabs

Werlabs is a registered Swedish healthcare provider built around prevention rather than treatment: regular blood testing, medical interpretation by doctors, and results presented over time in a digital journal. It works with accredited labs including Synlab, Unilabs and Karolinska Universitetslaboratoriet, and its medical team has analyzed millions of blood markers since 2013.

I joined as a consultant in 2024 and as an employee from February 2025, working on the consumer platform in an environment where correctness is regulated and a wrong result page is not a cosmetic bug.

Two services, owned end to end

The first was appointments and booking — the service that underpins the company’s paid video consultations. It integrates a third-party scheduling provider and handles entitlements, practitioner management, webhook consumers, Kafka events, and the booking and cancellation flows around them. I took it from design through to running it in production.

The second replaced something that kept breaking: a PDF generation microservice in TypeScript and Playwright, standing in for a fragile legacy JVM stack. Browser isolation, monitoring, and a path to autoscaling, because the failure mode of the old system was the kind that surfaces at the worst moment and nobody can explain afterward.

Around both: health-check coverage, structured logging, feature toggles, and deterministic proportional rollout. The unglamorous half — but a release you can turn off at 2% is a different kind of release.

Longitudinal medical data, drawn properly

I built a graphing library on Airbnb’s Visx for rendering medical results over time, used across both consumer and B2B surfaces. Large result sets, interactive time series, and a component API that other engineers build against.

The interesting constraint was not performance. It was that a person looking at their own values over three years should be able to see the trend without being alarmed by noise, and a doctor looking at the same chart should still see the outliers. Those are different readings of one picture.

Continuity through change

Over two years the team turned over several times, and I held the legacy platform through it — identity and user profile, results, the practitioner app — including handing a part of it to an external partner without dropping it.

I also introduced AI-assisted development to the team: editor tooling, shared rules, reusable skills, and the review habits that make them safe to use. The lesson was that adoption is not a tooling problem. People adopt what someone sits down and shows them, on their own code.

← All work