Hot pursuit after most recent AI trends. Maybe I could create neural network, then feed it with dozen of various pdf generated from DSLs and learn it to generate DSL having a PDF or image ?