Rag Enabled Clinical Decision System for Medical Recommendations

Medication safety remains a critical challenge in healthcare, as inappropriate prescriptions and drug–drug interactions contribute significantly 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. Quantization techniques 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.