Productizing ML Models with Dataflow – Towards Data Science

One of the key challenges I've faced in my data science career is translating findings from exploratory analysis into scalable models that can power products. In the game industry, I built several predictive models for identifying player churn, but it was always a struggle to get these models put into production. I've written about some of the processes used to productize models at Twitch, but each product team required a unique approach and different infrastructure. At Windfall, we're empowering our data science team to own the process for productizing models. Rather than relying on a engineering team to translate a model specification to a production system, we provide our data scientists with the tools needed to scale models.