For real breakthroughs in deep learning, we need a strong foundation. In this blog post, I would like to introduce Catalyst framework, developed with focus on reproducibility, fast experimentation and code/idea reusing. We’ll also provide a tutorial on MNIST classification problem as an example.