Memori
A voice-to-voice learning system for words, phrases, and real dialogue practice.
Memori turns language study into a live spoken interaction. The work centered on LiveKit research and a deliberately shaped realtime voice architecture with hooks into custom learning logic.
- Status
- Private client project
- Engagement
- Voice architecture and implementation
- Technical context
- LiveKit · Voice AI · Realtime audio · Custom business logic
The brief
Voice practice is not a generic assistant call. Each mode needs its own turn-taking, correction, repetition, progression, and recovery rules while still feeling like one uninterrupted conversation.
Engineering work
- 01
Evaluate LiveKit-based approaches for latency, interruption, session control, and observability.
- 02
Design distinct voice-to-voice modes for vocabulary, phrases, and dialogues.
- 03
Expose reliable hooks where product-specific learning logic can steer the realtime session.
Outcome
A private client system with a voice architecture shaped around the learning product rather than around a generic demo. The resulting interaction model and implementation were received enthusiastically by the client.
