This is a pattern I've had some success with several times now. Create a quick Python implementation for parts of data pipelines and then go back and re-write in Java/Go/C/whatever the best tool is for that bit of the job later when we know where the bottlenecks are.