Sleep Signal

Inspiration

My grandfather was a doctor. The system he devoted his life to failed him — his cancer was caught too late. If it had been found earlier, he would still be here.

That loss drove one question: where are the gaps the system isn't looking?

We found it. REM Sleep Behavior Disorder predicts Parkinson's with 80–90% accuracy — up to 14 years early. Almost nobody screens for it. The test costs $2,500. The window is being wasted.

What It Does

One night of bedside audio. A decade of early warning.

SleepSignal screens for RBD using just a phone — no hospital, no wearable, no cost. AI classifies sleep audio, estimates REM windows, and generates a transparent risk score. High-risk users are routed to neurologists and Parkinson's research trials — with full consent.

The twist: Pharma pays $25,000 per prodromal trial subject. We find them for ~$400. That's the business model.

How We Built It

Single HTML file. No build step. Runs in any browser instantly.

Two completely different interfaces unified by one toggle:

Patient View — warm, human, never scary
Clinical View — dark precision dashboard for clinicians and pharma

AI pipeline: Audio → YAMNet → LSTM Sleep Staging → Multi-class Classifier → Risk Engine

Challenges We Ran Into

The hardest problem wasn't technical — it was human.

How do you tell someone they might get Parkinson's in 10 years without devastating them? Every word and screen was designed around that question.

Technically: RBD sounds like sleep apnea. Distinguishing them is a genuine multi-class problem — not binary detection. We don't overclaim what a prototype proves.

Accomplishments That We're Proud Of

Dual design system — two completely different visual languages, one seamless toggle
Ethical disclosure flow that treats uncertainty with dignity
Found a business model hiding in plain sight — pharma funds what patients need free
Built a pharma recruitment dashboard most health startups never think to build