Wilsons-Navid/Capstone-Project

Rethicsec: AI-powered scam detection and one-tap cybercrime reporting for Africa. Flutter + Firebase app with custom multilingual scam classifiers (English, Portuguese, Swahili, Kinyarwanda). Two-stage: binary SMS model + four-class scan model (v3, macro-F1 0.932). BSc SE capstone, African Leadership University. Rethicsec AI-powered scam detection and cybercrime reporting for Africa | | | |---|---| | **Author** | Wilsons Navid Wado Tiwa, BSc Software Engineering, African Leadership University | | **Product** | Rethicsec mobile app (Flutter + Firebase + a custom Python ML scam classifier) | | **Status** | Implementation and Testing milestone. The deployed Android build is linked below. | There is no iOS or web build of the app. Full install steps are in section 1 . Just want to test the scam detection without installing anything? Use the live demo website and its model playground . Note: the demo site and model playground exist only to test the scam-detection models in the browser. They are not the app and do not represent the full product in any way, to actually use and experience Rethicsec, install it. **Rethicsec answers a simple question for everyday users: is this message a scam?** It gives a clear verdict in seconds and then helps the user act on it, including reporting the scam to the right national authority. The app uses custom-trained, multilingual scam classifiers rather than a general-purpose LLM, together with an education hub, an AI assistant, and an authority-reporting directory that covers 14 African countries. Detection runs in two stages: a fast binary "scam or not" model backs the SMS inbox feature, and a four-class model backs the manual scan, naming the scam type (advance-fee fraud, mobile-money fraud, phishing, or not-a-scam). Both are trained on real African scam messages in **English, Portuguese, Swahili, and Kinyarwanda**, the last from a live smishing honeynet run by the Upanzi Network at CMU-Africa. > **Why it matters: the reporting gap.** Cybercrime in Africa is badly under-reported. INTERPOL estimates > that fewer than 20% of incidents are ever formally logged, so the official statistics, and the > institutional response built on them, cover only a fraction of what actually happens. Rethicsec is built > to lower the ba …