EchoPanel
macOS menu bar audio capture & transcription
Problem
Professionals in meetings, interviews, and research need to capture and search audio content without switching between multiple tools or relying on cloud services.
Approach
Swift menu bar app using AVFoundation for audio capture. FastAPI backend with Whisper for transcription. SQLite for local transcript storage with full-text search. Designed for minimal UI presence with maximum utility.
Technical Implementation
Audio Capture
AVFoundation audio engine with custom audio buffer management. Captures system audio via CoreAudio APIs. Real-time streaming to backend.
Transcription
Whisper model (base/large) for on-device transcription. Chunked processing for real-time results. Speaker diarization experiments.
Storage & Search
SQLite with FTS5 for full-text search. Local-first architecture with export options (SRT, TXT, JSON).
UX Design
Menu bar widget with global hotkey. Minimal UI: record button, status indicator, transcript view. Searchable history with timestamp navigation.
Integration
FastAPI backend for optional cloud processing. WebSocket streaming for live transcription preview.
Outcomes
- →Native macOS UX with global hotkey activation
- →Real-time transcription with Whisper on Apple Silicon
- →Local-first storage with full-text search across transcripts
- →Export formats: SRT, plain text, structured JSON
Ownership & scope
Owned prototype architecture and implementation across native app layer, transcription pipeline integration, and local-first transcript retrieval flow.
Constraints
- →Needed low-friction capture from a menu bar workflow, not a heavy standalone app
- →Latency had to stay acceptable for near-real-time transcript feedback
- →Local-first storage and search expectations for privacy-sensitive usage
Trade-offs
- →Prioritized reliable capture + searchable transcript history over broad collaboration features
- →Used practical chunked processing strategy to balance speed and accuracy during live sessions
What changed
- →Audio capture and transcript retrieval moved from ad-hoc tools to one integrated flow
- →Users could search transcript history by content and timestamp instead of replaying long recordings
Workflow artifacts
- →Menu bar recorder with status and quick actions
- →Transcript history with full-text search
- →Export pipeline for SRT/TXT/JSON handoff
Result
macOS menu bar prototype combining native UX, audio capture, and AI transcription.