Project case study · AI / browser automation
DarkSense AI
An exploration of detecting potentially manipulative interface patterns on websites using browser automation, text classification, visual analysis, and heuristic signals.
Problem
Dark patterns are interface choices that can steer users toward actions they might not otherwise choose. Detecting them is difficult because relevant signals can appear in text, layout, and interaction flows rather than in one isolated element.
Approach
- Use browser automation to inspect web interfaces.
- Explore text classification signals, including a BERT-based approach.
- Consider visual signals and heuristic rules alongside text.
- Keep implementation evidence and evaluation separate from claims about model performance.
The repository includes the project interface and an implementation flowchart. See the source for the current implementation details.
Evaluation and limitations
Detection quality depends on representative labelled examples, clear category definitions, and evaluation that avoids leakage between training and test data. This page does not claim a specific accuracy, precision, recall, dataset size, or production readiness without a public, reproducible evaluation report.
Project status
The project is ongoing. The source repository is the authoritative place to inspect the current implementation.