Building and deploying a machine learning model for a FinTech-product.
Lightning talk - in Norwegian
Buffer is a FinTech-product that give companies loan in seconds. It connects to cloud-based accounting system,
analyzes invoices and give a suitable credit that changes with the activity of the company.
The decision engine has been developed in R, a language which historically was not suited for production, but the rise of container technology has changed this. We will also talk about how a close collaboration between e.g. developers, data scientists and UX-designers has made the development process lean.
In this talk we will take a peek under the hood, and look at the models that is behind the product. We have developed
a state-of-the-art machine learning model that predicts the probability of an invoice being paid, using the popular XGBoost-algorithm. We will cover how it was developed, what challenges we met and what pitfalls that should be avoided during model development.