FreeStyle Libre
Ten years on the part that cannot be late. In 2014 a sensor launched that let people see their glucose without a fingerstick. We joined the team in 2016 and took one component. We led the alarms: the part where arriving late and being wrong are the same failure.
You've heard of it. Here it is, through the public record.
Confidentiality is part of the job in medtech, so we describe the platform only in facts it has published about itself. It won the Prix Galien, often called the Nobel Prize of medical technology, for best medical technology. It was later named the best medical technology of the last fifty years, chosen from every Prix Galien winner since 1970. Its user base went from a few thousand to more than six million people across sixty countries. Its maker has publicly called it the most successful medical device in history in revenue terms.
We didn't do any of that. A very large and very good organisation did, over more than a decade, with hundreds of people. What we did was show up in 2016, take one component, and not leave.
Named best medical technology of the last fifty years
Galien FoundationPeople relying on the platform, across sixty countries
Public figuresDescribed by its maker as the most successful medical device in history by revenue
Public statementYears we have led the alarms
2016 — presentIn a monitoring product, the alarm is the product.
Everything else the software does, the graph, the trend arrow, the fourteen-day history, is context. The moment that changes an outcome is the one where the system decides something is wrong and tells a human being about it.
That moment happens on consumer hardware nobody controls, over a radio link that drops, on an operating system whose entire design philosophy is to stop your app from running in order to save battery. iOS is not hostile to this use case. It is indifferent. That is harder, because indifference gives you nothing to argue with.
Nobody has ever left a five-star review because a notification arrived on time. That asymmetry is the whole character of the work: you are engineering something whose success is invisible and whose failure is the only thing anyone will remember.
An alert that arrives late is not a degraded feature. In this product it is a wrong answer.
Ordinary app engineering treats latency as a gradient: faster is better, slower is tolerable. Alarm engineering doesn't work that way. There is a window inside which the alert is useful and outside which it is noise. The window is not negotiable, because human physiology set it, not a product roadmap.
So the work is not "make it fast". It is know whether it arrived, in every state the operating system can put you in: foreground, background, suspended, terminated, device asleep, radio dropped, phone rebooted, permissions revoked and re-granted at three in the morning. Each is a separate code path with separate guarantees, and the component has to behave identically from the user's point of view across all of them.
Ten years also means ten years of the ground moving. Every major OS release rewrites some assumption about background execution. The component has survived all of them, which is a different skill from building it in the first place, and the more valuable one. That regression, run by hand on a desk of phones for a decade, is what became Reach.
A naïve implementation looks perfect in the two states a developer tests in, and fails in the four states real people spend their lives in. Almost all of the engineering lives on the right-hand side of this chart, and none of it is visible in a demo.
Embedded work is rarely one clean deliverable.
We moved wherever the team needed depth. That is the honest description of what an embedded pair is for, and over a decade it adds up to something closer to institutional memory than to a contract.
Joined the team
Two years after the platform launched. We came in as embedded engineers and stayed embedded.
The alarms component
Ours to lead: delivery through every OS state, alarm logic, escalation, de-duplication, and the verification evidence that goes with a regulated release.
Implementing the design
Working directly with the design lead to get the interface built as intended — including the parts where medical constraints and design intent have to be negotiated line by line.
Wherever depth was needed
Watch prototyping for the innovation group, Android engineering, and the internal tooling that makes a regulated release process survivable.
Alarms
The component that carries the safety burden. Delivery through every OS state, under intermittent connectivity, with the evidence trail a regulated release demands.
Interface implementation
Building the design as designed. In a medical product that means negotiating pixel intent against clinical constraint constantly, and knowing which one wins in which situation.
Apple Watch prototyping
Exploratory work on what a wrist-first experience could be for continuous monitoring, where a glance replaces an unlock and the constraints get tighter, not looser.
Android and internal tools
The same class of problem on a platform with a different execution model and a fragmented device landscape, plus the unglamorous software the teams needed and never had time to build.
Ten years of alarms, written down and run by a machine.
The regression we ran by hand on every iOS beta is now the matrix. The seat we held inside the team is now a mode. Reach is this decade, packaged so that the next pump, pen or sensor does not have to learn it the hard way.
The matrix, on every OS beta.
Focus modes, deep sleep, permission resets, accessories, reconnection, interruption levels, receipts. Run on real devices against your build, with a report you can file.
- Kit: open source, in development
- Report: public, one per beta, in preparation
- Delivered, displayed, acknowledged, with timestamps
The same seat, in your repo.
OS-beta compatibility, alert regression, Bluetooth stack changes, SBOM refresh and vulnerability triage, delivered as a monthly compliance report under your quality system.
- Available now
- Your process, your definition of done
- A name on every line
We did not build the product. We were handed the part where late and wrong are the same failure, and we are still holding it.