Ronquy — On-Device Snore Detection
A private mobile app that detects snoring on-device, overnight, with no audio leaving the phone — a YAMNet + TFLite model runs an 8-hour audio-and-inference loop locally, with an optional cloud account for sync.
Business Context
Privacy is the product. A snore tracker that never uploads your sleep audio is fundamentally more trustworthy than one that does — and doing the inference on-device also means it works offline and costs nothing per night to run.
Strategic Value
Ronquy proves a hard mobile-ML pattern end to end: a native, battery-aware, all-night audio+inference loop with a real model on-device, wrapped in a cross-platform app and backed by an optional own-auth cloud. It is a template for any privacy-first, on-device sensing product.
The Challenge
Sleep-audio analysis is intrusive by default: most apps stream or upload recordings of you sleeping. Doing the detection honestly means running the model on the phone, all night, without draining the battery or sending audio anywhere — a hard constraint for a JavaScript app.
Our Approach
Ronquy runs a real on-device snore-detection model (YAMNet via react-native-fast-tflite) inside a native overnight loop that captures audio and runs inference locally for ~8 hours. The heavy, time-critical path is native rather than a JS shim. Audio never leaves the device; only derived events are kept. An optional cloud mode (FastAPI + Postgres + own auth) lets a user register and sync results across devices, but the detection itself is fully local.
Key Performance Indicators
| KPI | Baseline | Result | Impact |
|---|---|---|---|
| Privacy of sleep audio | Streamed / uploaded | Stays on-device; only events kept | Trustworthy by construction |
| Detection model | Cloud inference / mock | Real YAMNet + TFLite, native loop | Works offline, all night |
Proprietary — source code not publicly available
Architecture
ronquy architecture
Privacy is the product
Ronquy is a private mobile app that detects snoring on-device, all night, with no audio ever leaving the phone. A real YAMNet model runs through TFLite in a native overnight loop; only derived events are stored. An optional cloud account adds cross-device sync, but the detection itself is fully local.
This is proprietary work; the app is private. The card describes the architecture and intent without exposing internal logic.
Technology Stack
Visual assets for this project are not publicly available.
This is a proprietary project. Source code and external resources are not publicly available.