Andrew Dyuzhov

I build MetaWhisp — a free voice-to-text app for Mac that runs Whisper large-v3-turbo entirely on-device via Apple's Neural Engine. No cloud, no subscription, no audio ever leaves your laptop. Pro tier adds optional cloud transcription and AI text cleanup.

Before MetaWhisp, I spent three years and $2,400 on cloud transcription APIs producing podcasts. The bills kept climbing and I kept worrying about who actually saw the audio. In late 2024, when Apple's WhisperKit landed and the Neural Engine on M-series chips became fast enough to run Whisper large-v3-turbo at real time, the economics of paying $0.006 per minute to a third-party cloud collapsed. I shipped MetaWhisp four weeks later.

Since launch, MetaWhisp has processed ~18,000 hours of audio entirely on user devices. Zero cloud uploads. Zero data egress. Most users I hear from are lawyers (attorney-client privilege), therapists (HIPAA), journalists (source protection), and developers dictating into Claude or Cursor. Hard to overstate how many of them just wanted a tool that physically can't leak.

I work solo. My pair-programmer is Anthropic's Claude on the $200/month Max plan — she writes the code I dictate, ships the daily blog articles, reviews my CSS, and never asks for equity. The setup deliberately eliminates the coordination overhead of a small team. Product decisions ship the day they're made.

Previously, I led growth at Selzy (MarTech), where I learned that the fastest way to build something useful is to build the thing you already needed yourself.