Wearable Edge-AI Assistant for Multilingual Safety Communication on Construction Sites: A Heat Stress Prevention Application

Language discordance remains a persistent barrier to effective safety communication on U.S. construction sites, particularly for time-critical hazards such as heat stress. Existing digital safety tools and translation applications rely on scripted content, text-based interfaces, or cloud connectivity, limiting their effectiveness in dynamic, multilingual job-site environments. This paper presents a wearable, fully offline edge-AI assistant for real-time bilingual safety communication, with a specific focus on heat-stress prevention. The proposed system integrates offline speech recognition and lightweight LLM-based bilingual translation on a low-power NVIDIA Jetson Nano platform. To ensure fidelity to safety-critical terminology, an 80-sentence English-Spanish heat-stress dataset was developed from OSHA-aligned safety guidance. System performance was evaluated through a real-time feasibility demonstration and offline benchmarking. This work represents the first wearable, edge-only bilingual safety assistant demonstrated for feasibility in heat-stress prevention scenarios and designed for construction jobsite use.