A health-data lakehouse that turns messy FHIR, claims, wearable, and notes data into three trusted outputs — analytics marts, an ML feature store, and a vector index — with de-identification, OMOP standardization, and cross-engine parity testing that caught a real de-id gap.