CrisisSignal: Voice-to-Command Emergency

CrisisSignal is an English-language, multi-hazard emergency operations dashboard built around real places in Vinh City, Vietnam. It simulates multiple concurrent emergency calls for floods, tropical storms, and landslides using natural two-voice conversations generated in advance with Gradium. As each call plays, its audio is streamed progressively into AssemblyAI Universal-3 Pro Streaming, allowing the interface to reveal finalized speech in real time instead of displaying a prewritten transcript. AssemblyAI’s LLM Gateway converts the growing transcript into structured incident memory, confidence-aware facts, operational analysis, and recommended follow-up actions. Extracted locations are geocoded dynamically and displayed on a shared MapLibre map with incident metadata and measurements such as water level. Operators can monitor multiple calls, review the latest AI analysis, identify affected people and locations, and confirm simulated actions such as notifying disaster-prevention teams, dispatching rescue canoes, closing unsafe routes, requesting utility isolation, or issuing public warnings. CrisisSignal demonstrates how voice AI can reduce information loss and speed coordination during high-pressure, multi-hazard response. It is a hackathon prototype: all calls, incidents, measurements, and dispatches are simulated and the application does not contact real emergency services.