Not to be too self-promotion-y, but I work on an open source ML deployment that we built specifically because of how incongruous the data science workflow is to software engineering: https://github.com/cortexlabs/cortex