PPG · EEG · sleep-state classification · blood-pressure estimation · public prototypes

Pipeline diagram showing PPG and EEG signals, windows and features, model step, and sleep, ABP, and CAP outputs
Project boundary: physiological signals into experimental ML outputs, with clinical-use claims intentionally out of scope.

This page groups older physiological-signal ML projects. They are useful as evidence of applied ML range, but they should not be read as current clinical software. I am keeping that boundary explicit because health-adjacent ML needs tighter claims than ordinary portfolio copy.

How to read this page

  • PPG to ABP: waveform-estimation prototype around photoplethysmography signals.
  • Sleep/awake classification: PPG feature classification for state prediction experiments.
  • EEG CAP analysis: older EEG analysis work grouped with the same signal-ML track.
  • Boundary: research prototypes only; no clinical, diagnostic, or medical-device claim.

The shared signal pipeline

The common workflow is signal windowing, feature construction, model training, and evaluation against a task-specific output. The details differ between PPG and EEG, but the engineering problem is similar: noisy physiological time series need careful preprocessing before a model result means much.

This is why I keep the project line separate from current research papers. These repos show earlier applied ML range, not a current claim that the models should be used for decisions about someone's health.

public work

related writing

What held up and what I would change now

The part that held up is the decomposition: start with signal windows and features before making model claims. The part I would change today is the evidence bar. For health-adjacent ML, I would document data provenance, splits, leakage checks, calibration, error bands, and intended non-use more explicitly before publishing a polished project page.

Project artifacts