KYKirlos Yousef
All work

Independent product · Voice intelligence

Revera — AI Note Taker

A privacy-minded voice recorder and AI transcription product with offline queuing, interruption recovery, summaries, translation, and searchable notes.

Role
Creator & iOS Engineer
Period
2025 — Present
Platforms
iOS · iPadOS
Revera app icon with a white waveform in a circle on black
Contribution

Built the recording, transcription, persistence, recovery, and storage pipeline end to end.

~82%

less raw upload data

Offline-first

recording and queueing

Product context

Revera turns meetings, lectures, interviews, and voice notes into transcripts and structured information. Its experience is designed around an important reality: recording must remain trustworthy even when the network is not.

Engineering approach

The recording pipeline handles interruptions and queues transcription work until connectivity returns. Audio is prepared for speech processing by converting stereo 44.1 kHz input to 16 kHz mono. SwiftData keeps recordings, transcripts, and recovery state available locally, including explicit responses to low-storage conditions.

What the work demonstrates

Revera combines audio engineering, local persistence, applied AI, offline product behavior, and careful handling of failure states in a user-facing product.

Engineering challenges

  1. Protect long recordings from interruptions, low storage, and unreliable connectivity.
  2. Reduce audio transfer size without compromising speech-oriented transcription input.
  3. Keep recordings and transcripts recoverable through partial processing states.

Key decisions

  1. Combined SFSpeech and Whisper in a pipeline with offline queuing and interruption recovery.
  2. Converted 44.1 kHz stereo input to 16 kHz mono, reducing raw upload data by approximately 82%.
  3. Persisted recordings and transcript state with SwiftData and explicit low-storage handling.

Next case study

Plot My Calories