Jev maps Jack Cheng's speech and pointing to canvas editor calls

Result

Speech and pointing become one canvas editor call, or no change

Ask Jev which canvas action, target, and place to use from speech and a pointing finger.

The tweet is titled Jev is the future. The video is a canvas editor with three debug panes: INPUT, JEV, and CANVAS. INPUT holds speech, a pointing-word count, finger state, and the canvas as text. JEV returns action, target, and where, each with a percentage. CANVAS shows what the editor did and which API ran. A Calibrate button and a Listening label sit under the webcam. The overlay counts 7 or 8 questions. The tweet does not show the request.

The spoken INPUT on the film is draw a yellow circle here, make a blue square here, add a red diamond here, make it bigger, move, undo, select all, delete. The filmed JEV actions are clear, create, restyle, move, undo, and select. create paints a yellow circle, a blue square, and a red diamond. restyle makes the red diamond larger. move places the red diamond, the yellow ellipse, and the blue square, including to the right of the red diamond. undo reverses the last change. select marks three shapes. clear on an empty canvas does nothing in that region.

The canvas pane prints editor.createShape, editor.updateShape, editor.undo, and editor.setSelectedShapes. The overlay also prints millisecond figures such as 167 ms and 487 ms. Those are per-call traces, not a published median, so the metric row stays empty.