A $30 bedside device that spots sepsis hours before a nurse could.
Patient monitoring Unit for Low-resource Sepsis Early warning
The problem
In a Pakistani government hospital's general ward, one nurse can have 20 to 30 patients. Vitals are checked by hand every 4 to 6 hours. Sepsis can go from "just a fever" to organ failure in 3 to 4 hours, so patients get worse in the gaps between checks, and no alarm goes off.
Every existing early-warning system is built for the ICU and costs $50,000 or more. The wards with the highest risk have nothing at all.
What PULSE is
- Three sensors, no wires to a network. SpO₂, heart rate and temperature are read continuously at the bedside.
- A tiny ML model on the device. A quantized model reads a rolling 6-hour window of vitals on an ESP32-S3 microcontroller, completely offline.
- Alerts a nurse can act on. Green means stable. Yellow means recheck in 15 minutes. Red means call a doctor now, with protocol reminders on screen.
Why it matters
Every hour sepsis is caught earlier improves survival by about 7%. Equipping a whole 50-bed ward would cost about $1,500, less than a single commercial monitor. PULSE isn't a cheaper version of what exists. For most hospitals, it would be the first monitoring they've ever had.
Where it is
Stage 1 is under way: collecting data and training the first ML models. After that comes quantizing the model to fit on the chip, a hardware prototype, and a pilot at a government hospital in Multan, 10 minutes from my home.
Targets: AUROC ≥ 0.88 on just 3 vitals, about 4 hours of warning before clinical recognition, fewer than 2 false alarms per shift, and a model under 100 KB.
