At Careem, I developed and automated pipelines for training and evaluating dish retrieval embeddings, and created a unified dataset to support model development. I improved multilingual embeddings, increasing recall@10 by 10.2% and Arabic performance from 0 to 81%.
At Anghami & OSN+, I developed prediction and recommendation models, built data pipelines in Databricks with PySpark, and worked on Arabizi detection and machine translation. I also worked on semantic retrieval research and built benchmarks and dashboards to evaluate models.
