A complete voice pipeline
Global push-to-talk → audio capture → Whisper transcription → Spark → optional speech. Release handling, silence detection, and duplicate-send protection make the interaction usable.
PERSONAL PROJECT / VOICE-FIRST AI
Stay with the work.
Talk through the rest.
Talk to Spark to capture a task, move a meeting, or plan when to leave. I built this desktop assistant to keep my attention on the main task, with less reaching for the mouse and keyboard to manage everything around it.
ENGINEERING
Global push-to-talk → audio capture → Whisper transcription → Spark → optional speech. Release handling, silence detection, and duplicate-send protection make the interaction usable.
OpenAI tool calls map requests to task, calendar, and focus-session operations. A named Electron IPC bridge connects the React interface to services with context isolation.
Google Calendar OAuth supplies events and locations. A reminder scheduler combines upcoming commitments, task estimates, and starting-location context to suggest work that fits.
SQLite stores tasks and sessions with write-ahead logging and foreign keys. Plan, Focus, and Review views use the same records, including timers, subtasks, and notes.
LANGUAGES & FRAMEWORKS
Connected workflow: TickTick tasks, Google Calendar events, and Google Maps travel-time estimates bring task, trip, and departure planning into the conversation.