Categorizing Innovation.
My work spans five core pillars of modern AI, each designed to solve a specific class of global technical challenge.
Responsible AI & Governance
As Large Language Models (LLMs) integrate into society, making them culturally self-aware and ethically aligned is non-negotiable. My recent work focuses on mitigating algorithmic bias and ensuring AI understands the diverse cultural nuances of its users.
CALM: Culturally Self-Aware Language Models
Evaluating LLMs on Health-Related Claims Across Arabic Dialects
Multi-modal Learning at Scale
Real-world data isn't just text; it's images, signals, and structured metadata. I architect Hypergraph models that capture the complex relationships between diverse data types, currently operational at eBay and British Telecom.
Click-Through Rate Prediction with Multi-Modal Hypergraphs
Inferring Prototypes for Multi-Label Few-Shot Image Classification
Digital Health & Oculomics
Using the eye as a non-invasive window into entire human biology. My work in Oculomics combines computer vision with medical imaging to identify early biomarkers for cardiovascular and systemic health.
The Eye as a Window to Systemic Health: A Survey of Retinal Imaging
Would You Trust an AI Doctor? Building Reliable Medical Predictions
Probabilistic Foundations
The mathematical bedrock of my research. I develop Bayesian Non-parametric models and approximate inference techniques that allow AI to manage uncertainty in high-stakes domain-specific environments.