Jev Speech

An audio-first Socratic language-tutoring web app whose lessons stay on track with a TypeSafe/Jev Governor: System 1 evaluates every turn for topic drift, FSM state and Socratic compliance, never generating content. FastAPI backend, Next.js frontend.

An audio-native decision model shaped like Jev: audio is encoded once, several typed questions branch over that single encoding, and answers come back as calibrated probability distributions — no transcript, no ASR, the Whisper decoder never runs.

A study posing speech-neuroprosthesis rescoring as a single typed Jev decision. On 978 held-out sentences from a participant with ALS, Jev reaches 7.5% word error versus 7.8% for OPT-6.7b and Qwen2.5-7B, at USD 0.07 per thousand sentences with no GPU needed.