On-Device Audio Transcription

During technical interviews and screening calls, taking comprehensive manual notes while simultaneously engaging with interviewers causes distraction. Recruit includes a native, privacy-preserving audio pipeline built specifically for macOS.

Split Dual-Track Architecture

When you start an interview recording in Recruit, the system initiates two independent audio streams:

1. Candidate Mic Track (candidate_mic.wav): Captured from your primary microphone hardware with hardware DSP echo-cancellation.
2. Inbound System Audio (inbound_system.wav): Captured via Apple’s ScreenCaptureKit API directly from Zoom, Google Meet, Microsoft Teams, or WebRTC tabs without needing fragile loopback virtual drivers.

Both streams are written locally into the opportunity’s staging folder under ~/.recruit/recordings/<opp_id>/.

On-Device Speech Recognition (NPU)

Once the call ends, Recruit feeds the recordings into macOS’s native SFSpeechRecognizer running locally on the Apple Silicon Neural Engine (NPU):

• Zero Cloud Upload: Audio files never leave your Mac.
• Accurate Speaker Differentiation: Because the candidate mic and remote audio are tracked on separate channels, speaker attribution is 100% deterministic.
• Automated Debrief Synthesis: Recruit analyzes the combined transcript and extracts key technical questions, answers, and commitments into local_transcript.md.