Building resilient ML pipelines: Scaling data science impact with production-grade engineering