From prototype to product: 5 challenges and 5 advices about “productionizing” ML

The use of Machine Learning (ML) algorithms in products has been a reality only within the last few years. Hence, there is a limited experience about the challenges involved in designing and deploying a model.
These challenges include pipeline design, confidentiality, data and feedback gathering, periodic model retrain, ml-specific technical debt, performance/resource monitoring and continuous optimization, just to name a few.
Python is the de-facto tool for machine learning algorithms, but it also comes with several libraries to solve some of these issues, like flask, moto, pytest, ramlfications, airflow, luigi, etc
Hence, in this talk, I will present challenges and advices when developing products with machine learning.

Diego Alonso
Tech Lead, Synetiq

Diego holds a Telecommunication MSc from Spain and a Mathematical Modeling MSc from Denmark. Besides, he has 7+ years of professional experience in the design and implementation of solutions based on statistical models.
Previously, Diego worked as senior machine learning engineer at the Danish start-up Tradeshift developing the state-of-the-art system for retrieving info from purchasing documents (like invoices, credit notes) and a fully-python micro-service that asses similarity between entities (like companies, users,
products)
Now, Diego is working as tech lead at the Hungarian start-up Synetiq, where he contributes to build products based on video understanding using machine learning.