MLeap and Combust.ML allow machine learning pipelines developed in Spark to be deployed directly to production by serializing them to a common format and executing them outside of Spark. This addresses the common problem of data scientists developing models in Spark that then need to be rewritten by engineers for production. It also allows pipelines to be deployed via REST APIs with low latency. Benchmark tests showed average response times of 14ms for a linear regression pipeline and 24ms for a random forest pipeline on a MacBook Pro. Future work includes supporting more Spark and scikit-learn transformers and unifying model libraries with Spark.