This article describes a medical insights engine powered by artificial intelligence that utilizes natural language processing (NLP) and computer vision to decipher clinically meaningful content from unstructured medical transcriptions and radiological images. The system integrates biomedical Named Entity Recognition (NER), relation extraction, anatomical visualization, and temporal disease tracking using vision–language models. A Streamlit interface combines RAG-based clinical question answering and LLM-driven summarization. Experiments on medical transcripts, NIH ChestX-ray14, and biomedical ontologies achieved an F1-score of 0.91 for NER, 0.88 for relation extraction, and a BLEU score of 0.67 for image captioning. The framework demonstrates improved multimodal interpretability and unified clinical diagnostics.