Evidence
Methodology
The Marqi Index architecture: a 15-seed transformer ensemble with XGBoost out-of-fold meta-learner, built on structured EHR fields and closed claims data.
Architecture Overview
Full methodology documentation coming Day 5. Here's a preview of the technical approach.
Ensemble Architecture
15-seed transformer ensemble with XGBoost out-of-fold meta-learner. Multiple random seeds ensure robust feature learning.
Data Sources
Structured EHR fields and closed claims data. Hundreds of features engineered from demographics, diagnoses, procedures, and utilization patterns.
Retraining Cadence
Models retrained on rolling windows to maintain calibration as population characteristics evolve.
Validation Protocol
Temporal validation on held-out time periods. Calibration plot, intercept, slope, and Brier score reported.
Want the full methodology brief?
Schedule a 30-minute call to receive the detailed methodology documentation.
Request methodology brief