Our thesis aims to operationalise a machine learning workflow for a company or department that is interested to deploy a machine learning solution to enhance their work. The theoretical underpinnings of this machine learning workflow (also referred to as MLOps) will be based on materials stemming from different courses – ranging from infrastructural choices (e.g. Microsoft Azure, Databricks, AWS etc.), database choice (e.g. RDMBS vs NoSQL), database design theories, reference architecture and designing and crafting, to mathematical and statistical theories behind the machine learning algorithm that we eventually select for the solution, and other MLOps white papers or conference articles that explain best practices and recommendations.Current progress (research of start-ups in the space, industry practices and published peer reviewed papers) and monetary constraints has led us to narrow the project scope to operationalising an end-to-end ML pipeline merely using open-source software. Thus the thesis is converging towards "democratising" the deployment of MLOps in an inexpensive and simple manner.Research question:How to productionise MLOps in a live environment adhering to the principles of simplicity and affordability?

The value-add for the corporate collaborator is divided into two parts:Getting peer reviewed insights on best practices in academia and industry in terms of how to operationalise Machine Learning to be easily reproducable, partially automated and scaled (compared to stand-alone use-cases with e.g. Jupyter Notebook scripts)Recieving a practical (simplistic and affordable) demo case for a functional MLOps pipeline that usually takes months to construct with the help of costly consultants and expensive software subscriptions

The thesis topic is mainly motivated from current industry ML practices being very stale and hard to reproduce, scale and automate.Currently, Data Scientist’s and Data Engineer’s spend an overwhelming amount of time on manual data extraction to plug-in to static programming scripts to perform modelling to then finally obtain actual value from the vast pools of available data. Fortunately, implementing MLOps can bring this hassle to an end resulting in switching time spent and focus on actual ML-modelling and the resulting business value arising.

We have a well-rounded experience from driving corporate value from data (mainly working with Data Science at Novo Nordisk). Our experience spans further than this, with my expertise revolving around business strategy with past engagements in management consulting and investment banking, whilst Boon's expertise being in providing statistically robust insights in academia.In its essence, it is our combination of experiences from both the practical realms of programming (in ML, AI and NLP) and our capability to tie that technical knowledge to provide business value that is our most valuable asset for your company.