Edge Engineering for a Real-Time Smartwatch Telemetry Fleet
A life-critical health-monitoring system processing high-velocity telemetry (PPG, IMU, fall detection) from a fleet of battery-constrained smartwatches. Making it work meant bending the software to the hardware — from the wire protocol and on-device encryption to reliably commanding devices that are asleep 99% of the time.
The challenge
The watches barely had enough RAM to buffer a few minutes of telemetry, yet they were transmitting over HTTP — forcing the microcontrollers to construct bloated HTTP headers for every packet and pushing the hardware past its breaking point. Standard TLS/HTTPS was off the table too: the devices simply didn’t have the compute to handle the handshake. And because a watch must sleep 99% of the time to survive on its battery, sending commands back down to it is where most IoT architectures collapse.
Our approach
MQTT instead of HTTP
HTTP’s header overhead was choking device RAM. We ripped it out and moved the fleet to MQTT, whose microscopic 2-byte fixed header freed up the critical memory the microcontrollers needed.
Encryption that fits the hardware (ECC)
With the TLS handshake too heavy for the devices, we dropped connection-level encryption entirely, serialized the data into raw bytes, and encrypted each payload natively on the device using Elliptic Curve Cryptography.
Low-latency streaming backend
The byte-packed telemetry fed cleanly into a Kafka → Kafka Streams → Kubernetes pipeline, giving ultra-low-latency packet parsing and ML inference — fall detection and vital-sign anomalies — at scale.
Commanding sleeping devices (Redis + DCMS)
To reach devices that are almost always offline, a custom Device Configuration Management System wrote "desired states" and pending commands (SET_CONFIG, UPDATE) into a high-speed Redis cache keyed by device — decoupling cloud logic from the hardware’s sleep cycle.
The MQTT micro-window
Each watch subscribed to its own command topic. The millisecond it woke to publish vitals, the backend checked Redis for that Device ID and shot the binary command straight back down the open MQTT channel — configuration updated, back to sleep, all within seconds.
The outcome
The fleet stays asleep by default and transmits secure, byte-packed telemetry with none of the HTTP or TLS overhead the hardware couldn’t afford — while any watch can still be reconfigured or commanded within seconds of its next wake-up. Exactly what a life-critical, battery-constrained system needs.
Technology
Web standards are great for the web. But if you are building for the edge, your software must bend to the will of the hardware.