VERIDOSE

Infusion pumps are the most common device in intensive care — and the dose report comes from inside the pump itself. A kinked tube, a partial occlusion, a worn calibration, and the pump keeps logging volume that never left it. Independent verification exists in the lab; at the bedside, it's a graduated cylinder and patience.

I spoke with the clinical pharmacy team at a partner hospital: what they wanted wasn't another dashboard — it was a beep. Something that screams when the gap between expected and measured goes past 5%, with no network dependency, no screen, no login.

VERIDOSE is a small module that sits in the Set's path: a flow chamber with optical sensors, firmware in plain C (~3,300 lines, no RTOS), and a buzzer that exists for exactly one thing. All the safety logic is analog at the last instance — the emergency comparator works even if the firmware hangs.

I chose calibration with a pair of photointerrupters over a load cell: lower cost per bed and no strain gauge to recalibrate on every Set change. The firmware runs a bubble model (the classic false positive) and the alarm logic has double hysteresis — a silent early warning, and the audible alarm only after 10 seconds of confirmation.

The mistake that cost me the most: I read the sensor datasheet as “a reading every 1 ms” when it was a 4 ms moving average. In bench tests at low flow rates, underreporting hid inside the noise. It only showed up in validation with real fluid — exactly the kind of error a distilled-water bench hides. Since then, every validation runs on the protocol's clinical fluid.

In hospital pilot with 8 units, validated against reference gravimetry. Mean error of 1.8% across the clinical flow range. The next step is the compliance report for the regulatory process — the bureaucratic part is long, and it is the real deadline.