• Model development, training pipeline, and analytics backend • Maintain and improve the physics-based simulation engine — 19 equipment families, 64+ fault signatures, first-principles governing equations • Run model training pipelines — dataset generation, feature engineering, model fitting, hyperparameter tuning, MLflow experiment tracking • Implement model retraining triggers — drift detection (PSI-based), accuracy degradation monitoring, scheduled recalibration • Build and maintain the champion/challenger evaluation framework — shadow scoring, A/B testing, promotion guardrails • Develop new fault signatures as customer feedback identifies gaps • Implement probability calibration — Platt scaling, isotonic regression, ECE monitoring • Build the adaptive threshold controller — feedback-driven alarm threshold adjustment based on false alarm rate and recall • Develop the CMMS label linking pipeline — match work orders to predictions with confidence scoring • Analyze prediction outcomes — precision, recall, F1 by equipment family, by fault type, by site • Produce the weekly and monthly accuracy reports • Define and maintain feature sets for each equipment family — physics-informed features, rolling statistics, cross-tag correlations • Monitor data quality metrics — null rates, stale timestamps, schema violations, sensor drift • Build the healthy baseline update pipeline — daily computation of per-tag statistics from healthy operating data • Implement the training data snapshot pipeline — versioned, reproducible dataset extraction with manifest tracking