Rag Enabled Clinical Decision System for Medical Recommendations

V. Madhu, S. Sekhar, M. Jagan, M. Kavya ,P. Pradeep
Supervisor: Mr. N. YELLAJI RAO M.Tech (Ph.D), Assistant Professor, Dept. of CSE, VIET
Department of CSE , Visakha Institute of Engineering and Technology, Andhra Pradesh, India Department of
CSE , Visakha Institute of Engineering and Technology, Andhra Pradesh, India Department of CSE , Visakha
Institute of Engineering and Technology, Andhra Pradesh, India Department of CSE , Visakha Institute of Engineering and Technology, Andhra Pradesh, India

Abstract:Medication safety remains a critical challenge in healthcare, as inappropriate prescriptions and drug–drug interactions contributesignificantly to adverse events. Traditional Clinical Decision Support (CDS) systems are often rule-based and limited in adaptability, offering minimal personalization or interpretability. To address these limitations, this work Proposes a RAG Enabled Clinical Decision System For Medical Recommendations. The framework integrates patient symptom profiles and medical history with a neural backbone for medication prediction, enhanced through QLoRA fine-tuning to improve domain adaptation without extensive computational overhead. Quantizationtechniques are applied to enable efficient deployment on resource-constrained environments while maintaining performance. A deterministic safety module enforces drug–drug interaction and contraindication checks, and a Retrieval-Augmented Generation (RAG) layer grounds explanations in authoritative clinical guidelines and drug labels. A quantized large language model synthesizes these outputs into patient-friendly, disclaimer-aware explanations. Evaluation incorporates predictive metrics such as precision, recall, Jaccard similarity, and PRAUC, alongside safety indicators including drug–drug interaction rate and grounding accuracy.By combining QLoRA fine-tuning, quantized LLM inference, and RAG-grounded explanations within a CDS architecture, the proposed system advances toward safe, transparent, and computationally efficient drug recommendation. This work demonstrates how GenAI can be responsibly harnessed to improve clinical decision support while reducing risks associated with unsafe prescribing.1.1KeywordsDrug Recommendation, Clinical Decision Support, GenAI, QLoRA Fine-Tuning, Quantization, RAG, Medication Safety