AI Edge Disease Detection for Cold Environments

Inspiration

Winter conditions increase respiratory illness, and remote areas often lack connectivity for cloud-based health tools. We wanted a device that works anywhere, even in freezing environments, and detects illness before it becomes serious.

What it does

Listens to cough & breathing sounds Measures body surface temperature Uses a TinyML model to detect early signs of respiratory illness Works fully offline Syncs to a web dashboard when internet is available

How we built it

Trained a cough classification model using PyTorch Converted it to TensorFlow Lite Optimized using int8 quantization Embedded into an ESP32 / Raspberry Pi Built a React dashboard with real-time charts Used Node.js backend + Firebase for sync Performed feature extraction (FFT, Mel spectrogram) on-device

Challenges we ran into

Running ML on extremely low memory Noise reduction for outdoor cough detection Keeping device stable in cold conditions Ensuring secure data sync from the edge device

Accomplishments that we're proud of

Fully offline health monitoring Less than 300ms inference latency Model size reduced from 4MB → 350KB High accuracy on cough dataset (~86%) Dashboard with actionable health insights

What we learned

TinyML Hardware–software co-design (computer architecture tie-in) Optimizing models for edge inference Real-world health safety considerations

Built With

• github
• python
• react