Problem
Valuation modeling cannot stop at a single aggregate metric. Reviewers need segment-level performance, explainability, drift/performance signals, retraining decisions, and release-gate evidence.
/ Real Estate Valuation ML Governance
SPEC-NYC
A Python/TypeScript valuation-modeling and governance pipeline for NYC residential property data, focused on ETL, model evaluation, explainability, monitoring, retrain policy, and release-readiness artifacts.
System
Source connectors, canonicalization, ETL, identity/deduplication, feature engineering, time-based train/test splitting, XGBoost evaluation, SHAP artifacts, drift/performance monitoring, retrain-policy logic, champion/challenger code, Streamlit dashboard, and Next.js API/demo surfaces with Zod contracts.
Evidence
SHAPDrift monitoringRetrain policy
Stack
Python, Pandas, XGBoost, SHAP, Optuna, MLflow-style artifacts, Pandera, Streamlit, Next.js, TypeScript, Zod, PostgreSQL.