CursorAI
Predict user engagement from mouse movement patterns to enable early, actionable personalization.
Example output
Examples of engagement predictions generated from mouse movement data during the first seconds of a session.
What this project illustrates
-
Turning implicit behavioral signals into directly usable product insights
-
Designing lightweight, interpretable models for real-time inference
-
Balancing behavioral research with production constraints
-
Enabling early decision-making from weak but meaningful signals
Context & intent
Context
Traditional engagement metrics (clicks, conversions, time spent) appear late in the user journey.
This makes it difficult to react in real time or to understand user interest during the first moments of a session.
On many websites, users disengage before performing any explicit action.
​
Approach
CursorAI analyzes mouse movement trajectories as implicit behavioral signals.
By extracting spatial and temporal features from cursor dynamics, the model predicts user engagement within the first seconds of interaction—without relying on clicks or conversions.
​
Value
-
Predict engagement early in the session
-
Detect disengagement before explicit signals occur
-
Complement traditional analytics with behavioral insights
-
Enable real-time personalization and adaptive experiences
-
Improve understanding of user behavior beyond surface metrics
Project management & key decisions
-
Choosing behavioral features over content-dependent signals to ensure robustness across contexts
-
Favoring a lightweight and interpretable model rather than a complex deep learning approach, to support real-time inference
-
Focusing on short observation windows (first seconds of interaction) to maximize product value
-
Framing engagement as a binary, actionable signal rather than a complex behavioral taxonomy
Tech stack
-
Frontend: React, Netlify
-
Backend: FastAPI, Python
-
ML: XGBoost, behavioral feature engineering
-
Infrastructure: Docker, Google Cloud Storage, Cloud Run
Scientific background
CursorAI builds on work in cognitive science and human–computer interaction showing that mouse movements reflect attention, cognitive load, and user intent.
Cursor dynamics—such as speed, pauses, curvature, and spatial dispersion—provide implicit signals that can be leveraged to infer engagement before any explicit interaction occurs.