01 — Computer Vision · Flagship project
Predict visual attention before production.
Try the live demo↗A research-driven system that predicts where people are likely to look on an interface from a single screenshot — turning visual attention into an actionable design signal.
01 — THE CHALLENGE
Eye-tracking studies offer precise evidence, but they are costly and slow to run during everyday product cycles. The objective was to translate state-of-the-art attention research into an immediate, accessible UX tool.
Project anatomy
The system orchestrates asynchronous visual-saliency inference, then turns the prediction into a heatmap and contextual UX recommendations.
Core pipeline
Select a step to understand how the data is transformed.The user uploads a screenshot and defines the attention objective to evaluate.
Implemented features
Visible capabilities are connected to the technical components that make them possible.Upload of a web or application interface without prior instrumentation.
Execution container selection based on the available compute level.
The frontend polls job status and retrieves the result as soon as it is available.
Direct visualization of attention density over the analyzed screen.
The analysis accounts for the element the team actually wants users to notice.
Inputs, predictions and analyses remain organized around one shared identifier.
Stack by responsibility
Each technology is placed within its role in the system.Upload, processing status and visual output.
Request validation and job orchestration.
Image and result exchange across services.
Asynchronous CPU or GPU container execution.
SAM-ResNet, saliency losses and inference.
Heatmap composition and contextual UX interpretation.
The sequence below follows the product’s key moments: screenshot analysis, heatmap construction and interpretation of the areas most likely to attract attention.
Educational demonstration: data is local and some computations are simulated or accelerated.04 — OUTCOME AND IMPACT
“A production-oriented demonstrator that lets teams upload an interface and identify likely attention zones before conducting a live study.”