At Misraj, I lead the research team taking Arabic AI projects from idea to deployment. I architected and trained Kuwain 1.5B, an Arabic small language model built by injecting Arabic into an English-centric model while keeping its English ability.
I also built Baseer, a vision-language model for Arabic document-to-Markdown OCR, and Wasm, a pipeline for structured Arabic interleaved image-text corpora. Baseer reaches 0.25 WER on the Misraj-DocOCR benchmark, ahead of Azure AI and Gemini 2.5 Pro.
At EcoMundo, I led development of a pH forecasting model and built Graph-ML models for chemical challenge-test prediction, improving accuracy by 15% over previous methods. I also engineered data import pipelines that cut ingestion time by 40% on large datasets.
Earlier, at Omniya, I developed and deployed full-stack web applications that increased operational efficiency by 20%. I’ve co-authored seven papers and mentor a team of four junior researchers.
