Arabic Prompt Evaluation and Text Generation – Crowdgen (Fireweed Project)
Contributed to a large-scale LLM alignment and fine-tuning project focusing on Arabic. Tasks included evaluating AI-generated responses for relevance and accuracy, rewriting low-quality completions, and generating high-quality prompts and answers in both Modern Standard Arabic and Levantine dialect. Ensured adherence to quality guidelines, cultural appropriateness, and linguistic correctness. Provided structured feedback to support reinforcement learning from human feedback (RLHF). Worked with thousands of samples under tight deadlines while maintaining high QA scores.