The step from "it works in R&D!" to "hey, its deployed and we trust it!" takes a lot of work. I've been deploying working data science products from conception through to production for 17 years. I'll boil down the main steps you need to focus on to deal effectively with bad data, complicated data and algorithm dependencies, debugging, reporting and APIs and robust deployments. This talk focuses on Python but the main lessons and horror stories will apply to all languages.
If you're fresh out of academia and want to do data science then this will open your eyes to how 'stuff works in industry'. If you're in a growing data science team then you can learn from my mistakes! Be more effective, stop fighting fires and burning time.
