Multimodal CPAP / OSCAR analyzer — signal processing, phenotype classification, ML-driven intervention analysis.
CPAP Analyzer takes raw CPAP and OSCAR data and runs a multimodal analysis pipeline over it. Signal-processing layers extract respiratory rate variability, flow stability, and breath-detection events. A phenotype-classification layer maps those into sleep-apnea phenotypes, and an LLM-driven interpretation layer turns the result into something a person can read.
An ML intervention layer takes the next step: given a longitudinal CPAP history, it scores which interventions actually moved the needle — settings changes (pressure, EPR, ramp), mask swaps, lifestyle shifts — versus the ones that just looked like they should have. The goal is to make it possible to know whether a change is working in weeks, not months.
Privacy is baked in — the pipeline is designed to run locally on the user's machine, and no patient-identifying data leaves the device unless explicitly opted in.
Closely related to `oxyClean` (the earlier respiratory-rate-variability project) and `CoreSamples` (the HRV side from chest-strap data).