Cybersecurity · AI
ClearMessage
A research prototype that helps people understand suspicious messages and choose a next step. 2nd place at EVOHACKS 2026.
- Project status
- Research prototype
- Built by
- Hugo Kubíček
- Year
- 2026
Your parcel is waiting.
Pay now.
Why we built it
Suspicious messages can be hard to understand, especially for people less familiar with technology. We wanted to make the warning signs and the next step clearer.
What we made
The local application combines a trained text classifier, transparent rules, and practical guidance. Screenshot text recognition happens in the browser; confirmed text is analysed by the application backend.
The public demo replays a recorded result. It cannot analyse a visitor’s own messages.
TF-IDF and logistic regression achieved macro-F1 of 0.864 on a previously evaluated 880-message test set. This is an experimental dataset result, not independently measured performance on real-world fraud.
Next steps
Independent evaluation on more diverse messages, testing with real users, and a verified public deployment.