Shipping a transcription app to 300 users in two weeks
Cross-platform Flutter, real-time Whisper transcription, and the scrappy decisions that let a meeting app go from zero to 300 users without an ops team.
The brief was simple and the timeline was not: a cross-platform meeting app with live transcription, in two weeks. We hit 300+ users in that window. None of it was magic — it was picking the boring, fast path at every fork.
One codebase, two platforms
Flutter meant a single UI codebase for iOS and Android, which is the only reason a tiny team could ship both at once. We spent our saved time on the part users actually feel: latency.
Real-time transcription with Whisper
Audio is chunked and streamed to a Whisper-backed service; partial transcripts render optimistically and get reconciled as final segments land. The perceived speed comes from never making the user wait for a full sentence to commit.
- Optimistic partials keep the transcript alive while audio is still flowing.
- Reconnect logic assumes the network _will_ drop, because on mobile it does.
- A Gemini-based RAG prototype let users ask questions against past meetings.
Two weeks, one codebase, and a relentless bias toward the fastest correct thing.