Multilingual AI Tutor & Data Labeler at xAI
Project Scope: Contributed to the post-training and fine-tuning of a large-scale multilingual Large Language Model (LLM), focusing on improving reasoning capabilities in Arabic (MSA/Dialects), French, and English. Tasks Performed: Conducted Reinforcement Learning from Human Feedback (RLHF) by ranking model outputs based on Helpfulness, Honesty, and Harmlessness (HHH). Authored original "chain-of-thought" prompts to test logic and rewrote model responses to correct hallucinations and ensure cultural accuracy. Project Size: High-volume continuous workflow, processing approximately 400+ complex annotation tasks weekly within a distributed team of specialized tutors. Quality Measures: Adhered to strict, proprietary 50+ page style guides and Standard Operating Procedures (SOPs). consistently maintained a >98% accuracy rate on "Gold Set" benchmark tasks and achieved high inter-annotator agreement scores during consensus reviews.