DeepUX
Predict UX quality from a single screenshot using deep learning, enabling early design validation.
Example output
Examples of UX scores predicted from website screenshots across multiple UX dimensions.
What this project illustrates
-
Turning qualitative UX principles into quantitative, model-driven signals
-
Designing ML systems that approximate human judgment at scale
-
Applying deep learning to early-stage design validation
-
Bridging UX research, machine learning, and product decision-making
Context & intent
Context
UX evaluation is typically performed late in the design process, through user testing, audits, or A/B experiments.
These methods are time-consuming, costly, and difficult to scale—especially during early design iterations.
As a result, many design decisions are made without structured UX feedback.
​
Approach
DeepUX uses a convolutional neural network (CNN) trained on annotated website screenshots to predict UX scores directly from visual input.
The model evaluates an interface across multiple UX dimensions, approximating how users would perceive the design—without requiring any user interaction data.​
​
Value
-
Evaluate UX quality from a single screenshot
-
Get immediate feedback during early design phases
-
Compare multiple design variants objectively
-
Scale UX evaluation without user testing
-
Support UX, product, and design decision-making
Project management & key decisions
-
Frontend: React, Netlify
-
Backend: FastAPI, Python
-
ML: Convolutional Neural Network (CNN), TensorFlow
-
Inference: Vertex AI Custom Jobs, CUDA
-
Infrastructure: Docker, Artifact Registry, Google Cloud Storage
Tech stack
-
Frontend : React, Netlify
-
Backend : FastAPI, Python
-
ML : Keras, Vertex AI (Custom Jobs)
-
Infra : Docker, Artifact Registry, GCS
Scientific background
DeepUX builds on research in UX evaluation, cognitive psychology, and computer vision, showing that many usability and perception cues can be inferred directly from visual structure.
By learning patterns associated with human UX ratings, the model approximates user perception across multiple dimensions such as usability, clarity, and trustworthiness.