While there are many paradigms for scaling dev teams, this is not the case for data science teams.

In an attempt to keep up with company goals and growing demand for improved classification models, we had to shorten the research life cycle, and release the bottle neck of training, evaluating and deploying classification models. In this talk I’ll present the design of our model training platform - a set of independent tools and libraries, which allowed us to drastically increase the number of experiments we can perform, apply our findings easily to all existing models, and standardize the evaluation process.