Abstract
When disaster strikes, communication infrastructure fails precisely when it is needed most. Existing offline mesh networking tools enable device-to-device message propagation without internet connectivity but provide no mechanism to distinguish life-threatening signals from routine traffic-leaving rescue coordinators to manually triage hundreds of messages under extreme time pressure. This paper presents ResQMesh AI Platform, an open-source Android platform that integrates on-device machine learning into a Bluetooth Low Energy mesh communication stack for disaster response. Built on the BitChat BLE mesh protocol, the platform introduces four AI modules operating entirely without network connectivity or cloud dependency: a two-stage composite message classifier (M1), offline speech recognition (M2), an automated FEMA ICS-213 situation report generator (M3), and a batteryaware relay policy engine (M4). A BLE Priority Queue replaces FIFO packet ordering with priority-based preemption at the radio layer, reducing CRITICAL message delivery latency by three orders of magnitude under simulated mesh load. The M1 classifier combines approximately 90 deterministic FEMA/ICS keyword rules with a quantized Conv1D TensorFlow Lite neural model, classifying incoming messages into nine emergency categories across four priority tiers. The composite pipeline degrades gracefully to keyword-only mode on devices where the neural model asset is absent, ensuring full platform functionality across the complete spectrum of Android hardware. Classification results drive a real-time Emergency Feed UI and trigger automated ICS-213-compatible incident reports, providing rescue coordinators with structured, actionable information without access to external systems. The platform is released as open-source software under GPL-3.0, with all AI components isolated in a self-contained Gradle library module for independent testability and community extension. Evaluation methodology, dataset construction protocol, and on-device latency benchmarks are defined for reproducible comparative assessment. ResQMesh AI Platform demonstrates that semantic message classification, radio-layer emergency preemption, and FEMA-compatible reporting are achievable on commodity Android smartphones in fully offline disaster environments.

This publication has 6 references indexed in Scilit: