Pick a buyer (five industries, easy to hard, or write your own persona) and call them. Hold to talk, release to send; the buyer answers back in a synthesized voice and reacts to what you say. It's pure audio: no video, no transcript on screen while you're mid-call, just the pitch itself. Behind the scenes, Interhuman listens live over a direct WebSocket connection and surfaces signals and warning nudges as your pitch drifts: rushing past an objection, going quiet too long, losing clarity under pressure. When the call ends, a debrief screen scores the whole conversation: an overall score, talk ratio, pace, objections handled, and specific "watch out" moments timestamped against what was actually said. This native SwiftUI client talks to the same serverless backend as the web version (same personas, same live signals, same debrief), with one iOS-specific detail: the realtime endpoint rejects WAV but accepts m4a, so the recorder sends WAV to Whisper for transcription and a parallel m4a straight to Interhuman.

