CipherBridge

the problem

Industrial SCADA systems were designed for isolated networks, and it shows. Sensor telemetry moves in plaintext, devices trust whatever talks to them, and a single compromised node can feed poisoned readings to the control layer without anyone noticing until something physical breaks.

The constraint that makes this hard: you can't just bolt TLS onto an ESP32 polling sensors on a factory floor and call it done. Edge devices are resource-starved, latency budgets are tight, and encryption that slows telemetry past its usefulness is worse than no encryption at all.

what i built

A three-tier security framework that encrypts telemetry from the sensor to the cloud without blowing the latency budget:

• Edge (ESP32) — encrypts readings at the source before they touch the network, lightweight enough to run on microcontroller-class hardware.
• Gateway (Raspberry Pi) — aggregates edge traffic over MQTT, terminates and re-establishes encryption, and runs first-pass anomaly checks.
• Cloud (Flask + Redis + Docker) — ingests at 10k+ messages/sec and runs a dual-model AI engine: one model watches network behavior for malicious traffic (>88% detection), the other watches the readings themselves for tampered sensor data (>91% detection).

The split matters: catching a compromised network and catching a compromised sensor are different problems, and one model trying to do both did neither well.

how it works

End to end, a reading takes under 50ms from sensor to secured cloud storage: encrypted at the ESP32, relayed through the Pi gateway over MQTT, scored by both models, then persisted. Redis buffers the ingest path so throughput spikes don't back-pressure the edge.

what broke, what held

I attacked my own testbed six ways before trusting any number on this page.

results

• End-to-end encryption at under 50ms added latency
• >88% detection on malicious network traffic
• >91% detection on tampered sensor data
• Sustained 10k+ messages/sec ingest on the cloud tier