Mobile AI Assistant Offload Router

Every voice or text query a mobile AI assistant receives is first scored by a small on-device NLU model, then routed: answer locally right away, or ship it to a much bigger cloud model and wait for the round trip. This simulator runs that arbitration live — a confidence threshold decides the split, a network-RTT slider stands in for WiFi vs. weak cellular, and a dead-zone slider forces local-only fallback — while a 3D scene shows each query as a glowing sphere looping near the phone when it stays on-device or arcing out to a cloud node and back when it's offloaded. Live readouts track the device/cloud split, average end-to-end latency, cumulative battery drained, and the resulting average answer accuracy, so you can see exactly how the threshold trades speed and battery life against correctness — the same cascade Siri, Google Assistant and Alexa run behind the scenes on every command.