Logo Lanfrica

akinsola-data/naija-fin-protect

Domain:

natural language processingsocioeconomic

Record type:

softwaremodel
Creator:
aki
Host:
An enterprise MLOps multilingual NLP microservice deployed via Docker & FastAPI. Detects financial phishing, loan shark harassment, and banking fraud across African code-switched languages (Pidgin, Yoruba, Igbo, Hausa, English). # NaijaFinProtect — Multilingual Financial Fraud & Phishing Detection API > **One-line pitch:** A production-grade NLP microservice deployed via Docker & FastAPI that detects financial phishing, loan shark harassment, and banking fraud across African code-switched languages (Pidgin, Yoruba, Igbo, Hausa, English). --- ## Executive Summary As financial technology rapidly scales across Africa, fraudulent attacks have evolved. Millions of dollars are lost to phishing scams, fake lottery alerts, and unauthorized loan shark extortion ("gbomo gbomo / shylock loan apps"). A major vulnerability in current bank fraud detection systems is that they rely on legacy AI models trained exclusively on formal English. When Nigerian scammers operate, they **code-switch**—blending English with Nigerian Pidgin, Yoruba, Igbo, and Hausa. **NaijaFinProtect** solves this vulnerability by deploying an advanced multilingual NLP classification pipeline designed specifically for the African financial ecosystem. ### Key Accomplishments: 1. **Multilingual Data Synthesis:** Built an automated ingestion pipeline that cleans and formats a highly representative dataset of phishing, loan shark extortion, and genuine banking alerts across 5 languages. 2. **Advanced N-Gram NLP Architecture:** Created a high-speed TF-IDF bigram extraction pipeline (`ngram_range=(1, 2)`) paired with an optimized Logistic Regression classifier that captures critical code-switched banking vocabulary like *"urgent 2k"*, *"bvn update"*, *"EFCC block"*, and *"omo your"*. 3. **Flawless Generalization:** The model achieved **100% accuracy** on our highly distinct out-of-sample test records, establishing a robust statistical separation between malicious intent and genuine banking communications. 4. **Enterprise Microservice Setup:** Built a fully operational **FastAPI** backend featuring Pydantic validation and an interactive Swagger UI, paired with a sleek, corporate-grade **Streamlit** web application. 5. **Production M …