Overview
Built predictive audience demographic datasets using graph databases and AI/ML. Trained models to predict demographic attributes of mobile device audiences based on app usage, locations visited, and behavioral signals.
The core problem was inference at scale from weak signals: no single behavioral observation says much about who someone is, but the combination — which apps, which places, at which times — carries real predictive structure. Graph databases held the relationships between devices, apps, locations, and behaviors; the ML pipeline turned those relationships into demographic attribute predictions, packaged as audience datasets that advertisers and platforms could actually use.
This work sits in the Lumate/Great Data Lake lineage: the same discipline of turning raw behavioral exhaust into a governed, saleable data product, with the added responsibilities that prediction brings — being honest about confidence, and keeping the output inside the compliance envelope that Great Data Lake made a first-class part of the business.