New frameworks such as Spark, Tez, Flink, or Hive offer new possibilities, but using more than one framework is generally not easy. We build an abstraction layer that allows users to define their big data operations in a declarative way. This abstraction layer is backed by a platform that optimizes the workflow by carefully profiling each operation and running them, transparently, on the most appropriate framework given the description of the job and the currently available resources.