Ahmed Oumar El-Shangiti

About

I'm a PhD student in NLP at the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), supervised by Prof. Kentaro Inui. My research centers on mechanistic interpretability: probing and steering the internal representations of LLMs and vision-language models to understand numerical reasoning, counting, and the conditions under which model evidence collapses.

Alongside my doctoral work I am the founder of DataBayt.AI, building open-source software and datasets for the AI community. I collaborate with the RIKEN Advanced Intelligence Project, among other collaborators including Imperial College London and King's College London. I previously worked on generative AI for low-resource languages, machine translation, and large-scale dataset design.

Research Interests

Mechanistic interpretability Vision-language models Numerical reasoning & counting circuits Probing & activation steering Low-resource & Arabic NLP Multimodal reasoning

Selected Publications

Under
review

The Count Is There, but Misaligned: Understanding and Correcting Counting Failures in VLMs

Ahmed Oumar El-Shangiti et al., 2026 Under review

PDF

NAACL
2025

The Geometry of Numerical Reasoning: Language Models Compare Numeric Properties in Linear Subspaces

Ahmed Oumar El-Shangiti et al., NAACL 2025 Oral

ACL Anthology arXiv Code

ACL
2025

Palm: A Culturally Inclusive and Linguistically Diverse Dataset for Arabic LLMs

Alwajih et al., ACL 2025 Best Resource Paper Award

ACL Anthology arXiv Code

Arabic
NLP

Arabic Automatic Story Generation with Large Language Models

Ahmed Oumar El-Shangiti et al., ArabicNLP 2024 Oral

ACL Anthology arXiv Code

Preprint

Number Representations in LLMs: A Computational Parallel to Human Perception

AlQuabeh et al., 2025

arXiv Code

EMNLP
2024

Casablanca: Data and Models for Multidialectal Arabic Speech Recognition

Talafha et al., EMNLP 2024

ACL Anthology arXiv

EMNLP
2023

Dolphin: A Challenging and Diverse Benchmark for Arabic NLG

Nagoudi et al., Findings of EMNLP 2023

ACL Anthology arXiv

Invited Talks

Imperial College London

From detection to correction: activation-based error probes for VLM counting

In person · London, UK

Jun 30, 2026

Grammarly

Numerical reasoning in language models: a geometric perspective

In person · New York, USA

May 2025

UKP Lab, TU Darmstadt

The geometry of numerical reasoning in large language models

Online

Jan 30, 2025

Experience

Dec 2025 — Present

Founder & CEO

DataBayt.AI · Abu Dhabi, UAE

Leading development of open-source software and datasets for the AI community.

Jun 2025 — Aug 2025

Guest Researcher

UKP Lab, TU Darmstadt · Germany

Research on the interpretability of vision-language models (VLMs).

Aug 2024 — Present

NLP Research Fellow

RIKEN Advanced Intelligence Project · Japan

Interpretability of large language models.

Mar 2023 — Aug 2025

NLP Research Associate

MBZUAI · Abu Dhabi, UAE

Published at ACL, EMNLP, and NAACL on generative AI for low-resource languages, machine translation, and reasoning & interpretability in LLMs.

Oct 2022 — Feb 2023

NLP Research Assistant Intern

Qatar Computing Research Institute (QCRI)

Propaganda detection for Arabic and multilingual settings — 2nd place, SemEval 2023 Task 3.

Feb 2022 — Jun 2022

NLP Engineer Intern

Walmart Global Tech

Optimized production-scale LLMs for product-type prediction — 40–60% model-size reduction with maintained performance.

Education

2025 — Present

PhD Student, Natural Language Processing

MBZUAI · Abu Dhabi, UAE

2021

BSc, Mathematics & Computer Science

Cadi Ayyad University · Morocco

Awards

2025

Best Resource Paper Award

ACL 2025 — for Palm

2019

Facebook Secure & Private AI Scholarship

Meta / Udacity

2021

1st place — Arabic Misogyny Identification Shared Task

2023

2nd place — SemEval 2023 Task 3

Propaganda detection

Languages

Where I've Been

🇲🇷 Home 11 countries · 4 continents