Research
Exploring how foundation models can reason more reliably, represent diverse perspectives, and explain their decisions across languages and modalities.
Developing reinforcement learning methods that combine large language models with formal solvers such as Z3. This research studies whether symbolic verification can provide reliable process-level feedback, reduce reasoning errors, and improve generalization across logic and mathematical reasoning tasks.
Studying how a reasoning process can become inconsistent with problem constraints or with its own intermediate conclusions. The goal is to develop step-level methods that detect, explain, and repair these errors before they affect the final answer.
Designing preference learning methods that represent pluralistic, contextual, and culturally grounded judgments. This work explores hybrid verification, augmented intent modelling, and alternatives to evaluators based only on language models.
Investigating multilingual transfer, Arabic foundation models, knowledge editing, and interpretability across languages and modalities, with an emphasis on reliability beyond English-centric settings.
CONNECTED INTERESTS
RESEARCH FUNDING
The projects below distinguish awarded funding from projects recommended by Saudi Arabia’s Research, Development and Innovation Authority. Principal investigators are listed for each project.
Recommended by RDIA, Saudi Arabia
The project investigates data-driven methods for crowd analysis and management in cognitive cities, with applications to urban awareness, decision support, safety, and responsive city systems.
Recommended by RDIA, Saudi Arabia
The project focuses on Arabic large language models and their applications to language understanding, generation, reasoning, and domain-specific tasks.
Awarded · Defense Threat Reduction Agency, United States
The project developed methods for real-time infectious disease surveillance, geospatial monitoring, visual analytics, and coordinated response management in Pakistan.
Awarded · National ICT R&D Fund, Pakistan
The project investigated data-driven methods for dengue surveillance, epidemic modelling, visualization, and public health response management.