Research Prototype · Not a Medical Device

Your AI guardian for PTSD nightmares.

SentinelSleep listens while you sleep, detects nightmare acoustic signatures, and plays calibrated therapeutic audio — gently interrupting distress without waking you.

8M Americans affected< 3s detection latency0 wake-ups in demo sessions
82Recovery ScoreOptimal
3 Disturbances12m Awake85% Effective

May 1, 2026 · 7h 23m · 2 interventions

Why this matters

🧠
8,000,000

Americans experience PTSD-related nightmares each year.

💊
Drugs have side effects

Benzodiazepines and prazosin help some — not all. Image rehearsal therapy requires a therapist.

🛏
No one in the bedroom

Therapists and partners can't monitor every night. AI can — quietly, without disruption.

How it works

Four layers of protection

From audio detection to therapeutic response in under 3 seconds — running fully offline on your device.

01

Detection

MIT AST model classifies 2-second audio chunks across 527 AudioSet classes.

DSS > 0.4 → escalate
02

Verification

wav2vec2 returns valence, arousal, and dominance. Nightmare = low V + high A + low D, sustained ≥15s.

< 3s latency
03

Intervention

Pre-cached MusicGen + AudioLDM2 therapeutic audio plays at −20 dBFS, mixed with ambient sound for 60 seconds.

100% cache hit rate
04

Resolution

If distress resolves, system returns to listening. If it persists, a progressive wake protocol activates.

Morning report ready