Something I call a relational compiler. It makes Spark batch jobs run 10x faster by turning Postgres into a compute runtime.