Reads the expo-sensors Accelerometer at ~50fps and derives tilt magnitude and direction from the gravity vector with atan2. An adaptive low-pass filter (variable alpha plus a center deadzone) kills hand tremor while staying responsive, and the heading unwraps the 180/-180 seam so rotation never jumps. The smoothed values feed react-native-reanimated v4 shared values, with useDerivedValue, useAnimatedStyle, and useAnimatedProps driving the bubble offset, rotation, and degree text entirely on the UI thread. Glass bubbles use expo-glass-effect (iOS 26 Liquid Glass; Android falls back gracefully). Requires a dev build rather than Expo Go because of the native sensor and glass dependencies, and a real device since the accelerometer is inert in the simulator. Tested on Expo SDK 56.
A spirit level that reads the device accelerometer and spreads two Liquid Glass bubbles apart, rotating the pair toward the low side with a live degree readout behind them. Tilt the phone in any direction to see how far off level it is. Drop it into a measuring, DIY, or hardware-utility app.
A Telegram-style capture button whose microphone glyph morphs into a video camera on tap, with a springy pop under the finger and a liquid-glass circular surface. Every stroke is interpolated on the UI thread, so the mic cradle rises into the camera housing and the capsule head rounds into a lens in one continuous outline. Drop it into a chat composer, a recorder screen, or anywhere you need an audio/video capture toggle.
A pixel-matched recreation of Telegram's attachment menu: the paperclip button in the input bar morphs into a full liquid-glass bottom sheet, then drags between detents or flicks away to dismiss. Ships eight cross-faded panes (photo grid, text, file, location, poll, checklist, audio, contact) behind a scrolling native UISegmentedControl. Drop it into any RN/Expo chat screen that needs a real attachment picker.
A complete voice-assistant listening screen: a Skia orb that swells, fattens and ripples in time with whatever the microphone hears. Raw PCM is split into low, mid and high bands and pushed into Reanimated shared values, so the whole animation runs on the UI thread and React never re-renders while you talk. Drop it in as an AI-assistant, voice-note or call screen.
