Quantization for Efficient Enterprise AI Agents
As AI agents move from generating text to executing enterprise tasks, their computational requirements and energy consumption become practical deployment concerns. This talk explores how model size, quantization, and inference implementation influence task reliability, GPU memory requirements, and energy use. Drawing on experiments with tool-using AI agents in simulated enterprise workflows, it examines when compact models can deliver reliable performance and why reducing numerical precision does not always translate into energy savings. The discussion highlights practical considerations for building resource-efficient enterprise AI systems.
Speaker(s)
Dr. Prem Jagadeesan currently serves as the Head of Research and Development at Ducont. He is an Assistant Professor at the School of Artificial Intelligence, Amrita Vishwa Vidyapeetham, Coimbatore, and an Adjunct Professor in the Department of Health Sciences and Research at AIMS, Kochi. He earned his PhD in Data Science from IIT Madras and subsequently held a postdoctoral position at Purdue University. His research focuses on deep learning, model compression, agentic AI, and the geometry of AI models, examining how model structure and parameter sensitivity influence learning, reliability, and computational efficiency. He leads the AI-NEST lab, collaborating with students and industry partners to develop efficient AI systems for enterprise and healthcare applications.
