Challenge

The client’s product catalogue was extensive, while most of the meaningful data was unstructured, coming in the form of raw text or log events. Processing such data and training deep learning architecture with it required certain expertise. Taking into account that data drift is common in fast-moving environments such as this one, setting up an infrastructure capable of doing this fast and cost-effectively was needed.

Approach

  • Train a recommender system that merged data from different sources and formats.
  • Setting up a cloud-based infrastructure able to:
    • Process data and train a deep learning architecture on a schedule.
    • Ensure model monitoring.
    • Serve the recommender system as a containerised service accessible via API.

Result

  • Nucleoo provides a recommender system that is able to customise the products that are offered to web visitors, improving both visitor and lead conversion rates. Additionally, accessing system internals allows for customer/product clustering analysis.
  • The system is provided with its own CD/CI pipeline to allow easy deployment of new algorithm improvements as well as a model registry that enables model versioning and auditing.

Tech involved

AWS
Python
Docker
MlFlow
MySQL
TensorFlow

Value added

Nucleoo developed a scalable AI recommender system for the client’s online education platform, improving lead conversion rates and enabling customer/product clustering analysis.

Are you looking for a similar project?