My current research interests center on machine learning efficiency and multimodal perception, particularly how to make deep learning models small, fast, and accurate enough to run directly on power-constrained hardware. That includes model compression, quantization, and hardware-aware neural architecture search, along with multimodal systems that combine vision, audio, and sensor data into a single understanding of context. More recently, my focus has extended into agentic AI systems, including how autonomous agents are orchestrated and how to build safety and guardrail infrastructure around them.
Earlier in my career, my research was rooted in neuromorphic and event-based hardware: spiking neuron circuits, asynchronous sensing, and low-power analog design. That work still shapes how I approach efficiency today, even though my focus has moved well beyond it.
During my Ph.D. at the University of Florida's Computational NeuroEngineering Laboratory, my research focused on:
Low power neuron designs. Power dissipation constraints in analog circuits. Asynchronous analog-to-digital converters. Neural computation. Neuromorphic hardware systems. ACM/EKV models for subthreshold operation. Memristor technology.
This work formed the foundation for the patents and publications listed on my Publications page, including a Best Paper Award at IEEE ISCAS in 2011.