About

I am interested in AI problems that sit at the intersection of modeling and real-world use: systems that learn from feedback, make use of memory and personalization, reason across modalities, and stay useful under changing context.

That includes conversational and self-learning systems, retrieval and memory, multimodal and perceptual modeling, privacy-preserving machine learning, and AI for sensitive or regulated workflows where grounding, traceability, and reliability matter.

I am also drawn to questions of scalable inference and deployment, not as infrastructure for its own sake, but as part of what makes a modeling idea genuinely useful in practice.

My earlier work began in vision and perception: CloudCV, importance in images, and face detection and recognition systems. Those projects shaped how I think about contextual inference, ambiguity, and building models that have to function outside idealized benchmarks.

From there, my work expanded into conversational and self-learning systems, where memory, personalization, retrieval, and feedback become central to whether a model improves or slowly degrades in deployment.

More recently, that trajectory has extended into multimodal understanding, privacy-preserving generative AI, and healthcare workflows, where the technical problem is not only prediction quality but also grounding, traceability, and reliability in sensitive settings.