Snap Sleep is an agent-native, privacy-first, and self-evolving sleep health ecosystem. We are transforming multi-modal AI (vision and acoustics) into sensory infrastructure. Our AI agent doesn't just "diagnose" the mechanical causes of snoring—it actively "intervenes" by prescribing a customized hardware solution (our proprietary X-Wing module) and continuously self-optimizes its models based on real-world efficacy, creating an ever-improving Physical-Digital Recovery Loop.
Traditional sleep apps are passive dashboards. Snap Sleep is an Autonomous Health Agent.
- Sense (Multi-Modal Infrastructure): Utilizes edge computing (initially via iPhone, scaling to dedicated night vision cameras) to capture facial biomechanics and airflow acoustics in low-light environments.
- Diagnose: The AI engine accurately identifies the physical root causes of snoring (jaw drop, tongue collapse, palatal flutter).
- Intervene: The agent prescribes a specific physical intervention—our proprietary, rapid-3D-printable
X-Wingsilicone matrix. - Evolve: By measuring overnight biometric changes and subjective user feedback, the AI agent continuously trains and self-improves its diagnostic and treatment models.
During the POC phase, we strictly adhere to a "Zero Cloud Video Uploads" policy. All visual data extraction is processed in real-time on the local device (Edge AI) via Apple's built-in Neural Engine. As we scale, we only push de-identified numerical tags (e.g., timestamps, angle offsets, decibel parameters) to the cloud for deep model training. This ensures the highest privacy standards while empowering the AI to self-evolve.
- Visual Feature Engine (Vision Framework): Utilizes
VNDetectFaceLandmarksRequestto track facial dynamics in low light. Real-time calculation of the Y-axis distance between the Upper and Lower Lip accurately captures "Jaw Drop" events. - Acoustic Feature Engine (AVFoundation): Uses
AVAudioRecorder(isMeteringEnabled = true) to monitor the frequency and decibels (dB) of obstructed airflow, perfectly syncing with the visual timeline.
Converts the captured data into a structured 5W (What, When, Why, Where, Who) medical log and cross-references it:
| Medical/Anatomical Root Cause (The "Why") | Multi-Modal AI Logic (Vision + Sound) | AI Prescribed Intervention (The Hardware) |
|---|---|---|
| Mouth Wheezing | Extreme Jaw Drop Angle + Continuous dry airflow sound |
X-Wing Silicone Wings (Physically restricts oral turbulence) |
| Tongue Base Collapse | Lips closed or slightly parted + Sudden blockages / Choking |
X-Wing Internal Suction Bulb (Stabilizes and gently pulls tongue) |
| Palatal Snoring | Jaw slightly closed + Continuous low-frequency rumble (High dB) |
X-Wing Basic Kit (Alters oral cavity tension) |
| Nasal Snoring | Jaw Drop = 0 (Lips tightly closed) + High-frequency muffled hum |
Nasal Dilator Strip |
This is a revolution scaling from edge computing to a full hardware-enabled ecosystem:
- Phase 1 (Current) - Edge AI & MVP: Local multi-modal Proof-of-Concept based on iOS Vision & Audio Frameworks, paired with the Bambu Lab A1 rapid-prototyped X-Wing physical device.
- Phase 2 (Scaling) - Cloud Computing & Self-Evolving: Cloud integration for deep pattern analysis. Implementing User Feedback Loops to allow the AI agent to auto-train and optimize its recommendation models based on real overnight efficacy.
- Phase 3 (Ecosystem) - IoT & Night Vision: Launching the proprietary "Snap Sleep Night Vision Camera" with embedded machine vision and smart sleep peripherals, completing the self-improvement ecosystem for premium users.
This repository acts as a Showcase, designed to demonstrate the team's Agent-Native architectural capabilities and technical vision to investors and partners. To protect Snap Sleep's multi-modal core algorithms, self-evolving weights, and the physical parameters of the X-Wing, our core business logic is maintained as closed-source. We will gradually decouple and open-source our UI components, visualization funnels, and basic APIs to give back to the developer community.
Designed with 🌙 by the Snap Sleep Team. (Silicon Valley Software DNA x Taiwan Hardware Iteration)