Libre

When alerts fail, patients are at risk.

Abbott's FreeStyle Libre is a continuous glucose monitor worn by 4+ million diabetes patients. The system sends real-time readings to mobile apps and can alert patients and caregivers when glucose levels are dangerously high or low.

The existing alert infrastructure was showing cracks. During peak usage (mornings, meal times), alert delivery was delayed. Some caregiver notifications were failing silently. Sync between devices could lag by minutes—unacceptable for a patient safety system.

I embedded with the reliability engineering team. Mapped every data path from sensor to notification. Identified single points of failure and capacity bottlenecks.

Every millisecond matters.

Redesigned the real-time pipeline with patient safety as the primary constraint. Alert delivery is now a separate, prioritized path with its own capacity allocation and failover.

Architecture principles: no single points of failure for critical alerts, multi-channel delivery (push + SMS + email), automatic retry with escalation, and complete delivery confirmation tracking.

99.99% is not optional. It's the floor.

Implemented SRE practices with error budgets specifically designed for patient safety systems. The error budget for critical alerts is essentially zero—any missed alert triggers immediate incident response.

Built comprehensive observability: real-time dashboards for alert delivery, automatic anomaly detection, and on-call escalation for any delivery degradation.

When the alert fires, everyone knows what to do.

Created comprehensive runbooks for every failure mode. Trained on-call engineers on patient safety implications. Established clear escalation paths with medical team involvement for critical incidents.

Regular chaos engineering exercises: we deliberately introduce failures to verify that redundancy works and that the team responds correctly.

The Result

Alert delivery now runs at 99.99% with sub-second latency. Zero missed critical alerts since the new architecture launched. Caregiver notification reliability improved from 97% to 99.95%.

The platform now handles 4M+ patients with room to scale. The reliability improvements have directly contributed to better patient outcomes—caregivers trust the alerts, and patients trust the system.