AbdelRahim Elmadany

I build multilingual language and speech AI that works beyond the small set of high-resource languages that dominate today's models. My research connects large language models, speech, data, evaluation, and scalable machine learning to make AI more reliable in linguistically diverse, real-world settings.

Arabic and African languages are a central proving ground for this work. Their dialect diversity, code-switching, cultural context, and uneven data availability expose limitations that conventional benchmarks often miss. I develop models, datasets, benchmarks, and evaluation frameworks that help close these gaps and expand who modern AI can serve.

My goal is not only to publish strong research, but to turn it into reusable resources, open benchmarks, and systems that enable broader research and real-world adoption. This work has appeared at ACL, EMNLP, EACL, INTERSPEECH, LREC, and ACM CSCW, and is advanced through collaborations connecting research communities across Canada, Africa, and the Arab world.

My research develops multilingual AI systems end to end—from foundation models and speech technologies to the data, evaluation, and infrastructure required to make them reliable in low-resource and culturally diverse settings.