Commercial voice assistants are generic and disconnected from family context. They can't remember that Riley has a math test tomorrow or that Alex prefers Boston sports updates.
Built a local-first voice assistant that runs on distributed hardware (N100 mini PC → Raspberry Pi nodes), with per-family-member profiles and context awareness.
- Local-first architecture for privacy and speed
- Wake word detection via Picovoice
- ElevenLabs for natural-sounding TTS with voice cloning
- Deepgram for fast, accurate STT
- Family profiles stored locally, context injected per-request
Working prototype with touchscreen UI, multiple room support planned, and integration points for calendar, weather, and school schedules. Daily use for morning briefings.