A RESEARCH PROJECT ON AI CLASSIFICATION

Making Screen time classification less ambiguous using lightweight AI models

ClassifAI allows users to understand their screen time habits while having transparent and safety-first privacy.

Watch the demo

Raw Frame

Input

Inactivity → ResultActivity → Classify

Filter Active Content

Model A · 93% accuracy

WorkOREntertainment

Classify Intent

Model B · 86% accuracy

WorkEntertainmentNeutral

Work vs. Entertainment

Result

THE PROJECT

About ClassifAI

Our application determines how long a user spends on entertainment versus work. Screenshots pass through Model A, which classifies activity with 93% accuracy. If Model A identifies inactivity, the result is recorded. If it identifies active content, that information goes to Model B, which classifies the activity as work or entertainment with 86% accuracy. Once Model B finishes, the results are updated and all screenshots are deleted.

The models run on-device, and no information is saved externally or locally. Our goal is to make classifications clearer, communicate uncertainty, and help people make informed decisions. ClassifAI was created to help users become more aware of how they spend their screen time.

SEE THE PROJECT