Title: Artificial Intelligence Across Disciplines: Research, Innovation, and Intelligent Solutions
Editors: Dr. Subita Bhagat, Dr. A. Balamurugan, Dr. P. Krishna Kumar, and Mrs. S. Nandhini Devi
ISBN: 978-81-69857-83-3
Chapter: 4
DOI: https://doi.org/10.59646/815/04
Author: Dr. Suriya Begum
Abstract
Generative Artificial Intelligence and foundation models have transformed enterprise automation by exhibiting few-shot contextual reasoning and multi-modal generation capabilities. However, deploying multi-billion-parameter foundation models within high-reliability domains—such as clinical healthcare informatics—is impeded by ungrounded factual hallucinations, catastrophic inference latency, and computational costs during domain adaptation. This chapter proposes an end-to-end intelligent architecture that integrates Retrieval-Augmented Generation (RAG) using dense-sparse hybrid vector indexing, Parameter-Efficient Fine-Tuning via Low-Rank Adaptation (LoRA), and direct preference alignment. Using clinical diagnostic summary generation as an empirical case study across 25,000 multi-institution electronic health records, the architecture optimizes both parametric and non-parametric knowledge retrieval paths. Experimental evaluation demonstrates an 88.4% factual consistency score on clinical claim verification, a 71.3% reduction in hallucination frequencies relative to zero-shot foundation baselines, and a sub-350-millisecond token-generation latency achieved through hardware-aware memory kernel tiling, establishing a robust framework for mission-critical enterprise systems.
Keywords: Foundation Models, Generative Artificial Intelligence, Retrieval-Augmented Generation, Low-Rank Adaptation, Clinical Decision Support, Hallucination Mitigation.