Work
Each project has a write-up covering the problem, the data, the decisions I made and why, and what I would do differently. Status reflects what is built, not what is planned.
A machine learning system that reads an ECG and flags what warrants a clinician's attention. Version one detects atrial fibrillation from rhythm irregularity, built end to end from raw PhysioNet waveforms to a deployed application. The roadmap extends to ischemic ST changes, conduction abnormalities, and ECGs captured as photographs rather than digital signals.
Read the write-upHospitals now publish their negotiated rates. I matched those contracts across more than 1,700 US hospital price files to CDC Medicare data on who starts and finishes cardiac rehab. Commercial plans pay Washington hospitals the second-highest rate in the country, yet Washington enrolls and retains fewer patients than most of the US. The dashboard turns that gap into dollars for any state, and forecasts a program from its own monthly numbers.
Read the write-upThe ECG toolkit is at v1 and reads rhythm only. The next version adds image digitization, so a photograph of a paper ECG can be interpreted, along with morphology features that extend detection beyond rhythm to ischemia and conduction disease. Both are described in the write-up.