Mazin User

I’m an AI data evaluator focused on LLM assessment, RLHF feedback, and bilingual Arabic–English NLP analysis. I evaluate reasoning, instruction-following, semantic accuracy, and culturally sensitive outputs across platforms like Outlier, Toloka, and Prolific.

In my evaluator work, I apply rubric-based frameworks to judge coherence, factual accuracy, semantic consistency, and cultural alignment. I also deliver structured preference feedback and annotation rationales that support model fine-tuning workflows.

I’m particularly strong in Arabic dialect assessment, cross-lingual NLP quality, and AI alignment workflows, including safety & harm detection and prompt quality review. I maintain consistent reviewer quality scores against project benchmarks across long-term annotation and ranking tasks.

I also build my own benchmarks and documentation to improve evaluation rigor. My personal and independent research focuses on dialect coverage, error taxonomies, and actionable recommendations to strengthen Arabic training data quality—while I continue improving my analytical foundation through a B.Sc. in Civil Engineering.