At Roche, I proposed and led the rearchitecture of a batch Spark ETL platform into an Apache Iceberg lakehouse. Incremental CDC cut runtime from 12 hours to 40 minutes, lowered unit compute cost by 21%, and eliminated recurring production failures.
I also built and productionized a GenAI synthetic-data platform with healthcare-aware generation and privacy-risk auditing, keeping real PHI out of non-production environments. Earlier, at Cowbell Cyber, I owned an ingestion framework that scanned 1,000+ domains daily and cut cyber-insurance time-to-quote from five minutes to one.
