Recommender system

Our team developed a recommendation system for the staffing process that would ease business processes.

We built the system from scratch, starting from the data flow design, integration of specific business rules and restrictions into the product, and creation of an explainability layer for each model proposal.

The system was designed as a ranking algorithm built around a DNN with a custom multi-headed loss function in order to incorporate penalties from different types of environment feedback. Despite the tabular nature of the data, DL approach turned out to be the only one that could meet business process features. Due to recent changes in business processes, we also designed a solution for data drift detection and handling.