AI systems
HealthSense
A wearable-AI exploration that turns biosignals into an interactive dashboard, combining ECG classification, signal-processing heuristics, and an agent workflow.
HEALTHSENSEWEARABLE INTELLIGENCE
EEG · ECG · PPG · EMG · IMU
5 signals · ECG CNN + heuristicsSystem prototype
About the project
HealthSense brings ECG, EEG, PPG, EMG, and motion signals into one prototype interface. I implemented a small ECG CNN and ONNX export alongside signal-processing heuristics, then explored how an assistant could explain the resulting signals. The larger multimodal model and NPU performance figures are design targets; the work is exploratory and is not a clinical diagnostic tool.
What I built
- Built a prototype dashboard and agent pipeline for ECG, EEG, PPG, EMG, and IMU signals. Implemented signal-processing heuristics and a small PyTorch 1D ECG CNN with ONNX export.
- Explored a PanLUNA-inspired multimodal foundation-model design. The 5.4M-parameter architecture and 95 ms / 19 mJ NPU figures were concept targets, rather than measurements of a deployed model.
- Demonstrated the BioTrain adaptation workflow through simulated training curves; on-device backpropagation remained future work. ECG experiments were exploratory and were not subject-held-out clinical validation.
