Point of Care (POC) machines are deployed worldwide in laboratories, hospitals, and clinics to deliver fast test results right where care is given. Because these machines operate in dozens of countries, their screen text — instructions, warnings, status messages, and menu items — must be available in the right language for the user. For this manufacturer, keeping those translations up to date had historically been a slow, error-prone, and manual process.
In this project I delivered the integration between a visual translation system vendor and the manufacturer of these POC machines. The starting point was simple: let the translation team work in the tools they're great at — a visual translation environment with context, screenshots, and memory — and make sure the result flows back into the machines flawlessly.
The technical challenge centered on robustly and scalably parsing and transforming the XML screen definitions that drive the machines. That XML carries more than just text: layout hints, conditional visibility, IDs, length limits, and context that are essential for accurate translation. A translator who can't tell whether a button is part of an error dialog or a routine settings screen will inevitably make poor choices about tone and wording.
I built a conversion layer that turns this XML into the translation system's format while preserving all relevant metadata: context, maximum lengths, plural forms, variables, and screen placement. Translators see in their own interface exactly where a string lands in the machine, how much room is available, and which other languages are already complete — and they can draw on their existing translation memory.
On the other side of the pipeline I implemented a write-back path with extensive validation. Translations are checked for length, missing variables, disallowed characters, and consistency with previously approved translations. Only when all checks pass is the XML rebuilt and published back to the machines, so a bad translation can never cause a lab-floor incident.
The entire process is set up as an automated, repeatable pipeline. A new machine software version triggers a new translation round; completed translations flow back along the same route. That dramatically lowers the threshold for adding languages or refreshing existing ones, and for the first time gives the manufacturer real visibility into translation status per language and per screen.
What made this project valuable wasn't just the technology but the collaboration between two organisations with very different ways of working. By investing early in clear agreements about the XML contract, error handling, and translation ownership, the integration could keep delivering value even as both sides continued to evolve.
Outcomes
- One repeatable translation pipeline for machines operating in dozens of countries
- Higher productivity for the translation team by carrying context, screen placement and length limits along
- Validation on four points before publication: length, missing variables, disallowed characters and consistency
- First-time visibility into translation status per language and per screen