Research Themes

Categorizing Innovation.

My work spans five core pillars of modern AI, each designed to solve a specific class of global technical challenge.

Pillar I

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.

NeurIPS 2025

CALM: Culturally Self-Aware Language Models

ACL 2026

Evaluating LLMs on Health-Related Claims Across Arabic Dialects

Pillar II

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.

CIKM Best Paper Nominee

Click-Through Rate Prediction with Multi-Modal Hypergraphs

AAAI 2022

Inferring Prototypes for Multi-Label Few-Shot Image Classification

Pillar III

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.

Featured on Forbes

The Eye as a Window to Systemic Health: A Survey of Retinal Imaging

WISE 2024

Would You Trust an AI Doctor? Building Reliable Medical Predictions

Pillar IV

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.

ACL 2019

Word and Document Embedding with vMF-Mixture Priors

AAAI 2019

Word Embedding as Maximum A Posteriori Estimation

Full Publication History
Global Interest
Flag Counter