Ibrahim Abu Farha

I am an AI/NLP engineer with over six years of experience across natural language processing, large language models, speech technologies, and scalable AI systems. My work has always moved between research and engineering, and the thread running through it is simple: build language technology that works in the real world, especially for Arabic and the dialects that mainstream NLP tends to overlook.

I am currently a Senior AI/ML Engineer at OASES, part of TAG Software Group, where I drive applied AI across the group’s aviation portfolio. Day to day, that means building LLM-based multilingual assistants grounded in operational systems, evaluating computer vision models for ground operations, and applying machine learning to fleet reliability and maintenance risk. Before OASES, I worked on machine learning for industrial monitoring at Qualitrol, and I was the Founding AI/NLP Scientist at Alsun AI, where I built dialect-aware Arabic voice agents end to end, from speech recognition through retrieval-augmented generation to LLM-based dialogue.

On the research side, I was a Research Associate at the University of Sheffield, working with Kalina Bontcheva and Carolina Scarton on multilingual NLP for disinformation analysis as part of the EU-funded VIGILANT project.

In 2023, I completed my PhD in Informatics at the University of Edinburgh, supervised by Walid Magdy and Bonnie Webber, as a member of the SMASH research group. My thesis tackled Arabic sarcasm detection, a problem I care about because sarcasm quietly breaks sentiment analysis systems that otherwise look accurate. Along the way I built the ArSarcasm and ArSarcasm-v2 datasets and co-organised SemEval-2022 Task 6 (iSarcasmEval), the first shared task on intended sarcasm detection in English and Arabic.

Before the PhD, I completed my MSc in Artificial Intelligence at Edinburgh in 2018, and a Bachelor of Computer Systems Engineering at Birzeit University in Palestine, where I graduated as Valedictorian.

I am particularly interested in responsible and inclusive AI: systems that work for diverse languages, dialects, and domains, not just the well-resourced ones.