Owing to the rapid accumulation of other diverse data in the field of ophthalmology, including high resolution retinal image and clinical text records highlights the demand for high throughput analytical frameworks that are able to process and integrate these modalities. We present a novel ophthalmology-centered retrieval-augmented generation (RAG) framework, including architecture optimization via ResNet50 for images, BERT for texts and cross modal attention mechanisms for data fusions. Highlighting the improvements offered by this work over the state of the art, addressing issues around integration of multimodal data, scalability of the model and domain-specific optimization; this work marks an important step forward in improving the accuracy during the diagnostic process, ultimately aiding in better decision support systems. We validate our framework with empirical evaluation on the Suris+ dataset, showcasing that it consistently surpasses existing methods in retrieval precision, generation quality, and computational cost. These results underscore potential of the framework for real world clinical implementation.