LLM Fraud Detection

Detecting fraudulent activity in regional languages requires specialized NLP approaches. This project involved designing and training a dedicated fraud-detection model specifically targeting Arabic text.

Leveraging transformer-based Large Language Models (LLMs), I engineered a robust pipeline that included comprehensive data-cleaning and tokenization strategies suited for Arabic morphology.

A major focus of the project was building strict evaluation pipelines to ensure the model’s predictions were highly accurate and the results were fully reproducible for integration into future, production-level use cases.