RESCUE-Edge is Baraa Said's graduation project at Birzeit University: a disaster-resilient edge-AI framework that detects people and hazards room by room and keeps reporting over an authenticated LoRa link when the Internet is down.

LoRa radios reach across a building with very little power and need no network infrastructure. Ascon-AEAD128 is the NIST lightweight-cryptography standard, small enough for microcontrollers, so every frame can be encrypted and authenticated.

Is RESCUE-Edge finished?

The software foundation is built and tested, including the crypto, the rule engine, the offline assistant and the dashboard. The hardware build is in progress.

RESCUE-Edge: offline edge-AI safety sensing over LoRa

The problem

In a fire, an earthquake or a long blackout, responders need to know which rooms have people in them and which have hazards. That is exactly when a building's network and the Internet are most likely to be gone.

The design

Sensing. ESP32 sensor nodes and Wi-Fi motion sensing build room-level evidence of presence and hazards.
A link that survives. When Wi-Fi and the Internet fail, a LoRa radio link carries the evidence. Every frame is encrypted and authenticated with Ascon-AEAD128, the NIST lightweight-cryptography standard, with replay protection.
A command station on battery. A Raspberry Pi 5 runs the operator dashboard and an assistant with no cloud at all.
An offline assistant. It answers in Arabic or English from local documents, using hybrid retrieval (keywords plus embeddings) and a small language model running on the Pi through llama.cpp.

Built and tested so far

The radio wire format, the rule engine and the Ascon-AEAD128 implementation, verified against all 1,089 official test vectors.
165 automated tests passing on both a Mac and the Raspberry Pi.
The operator dashboard: ten screens in Arabic and English, running on replay data. Unknown or stale readings are drawn hatched, never green, so missing data can't look safe.

Honest limits

The hardware build is in progress. Images marked "concept render" are illustrations, not photos. Detection thresholds in the prototype are development values, not approved safety limits.