<p>Interpreting handwritten prescriptions remains a significant challenge in healthcare, often resulting in therapeutic errors and adverse patient outcomes. Although computerized prescription systems are increasingly adopted, handwritten prescriptions persist, particularly in resource-constrained settings. To mitigate this issue, we propose RxVLM, a Vision–Language Model (VLM)-inspired transformer-based framework designed for accurate and reliable handwritten prescription understanding. This framework incorporates data preprocessing, data augmentation, and a contrastive learning mechanism that effectively aligns visual and textual representations within a shared embedding space using cosine similarity and contrastive loss. A patch-based Vision Transformer encoder combined with Byte Pair Encoding (BPE) tokenization enables the model to capture complex lexical and visual structures inherent in handwritten medical documents. Experimental results on a curated dataset of annotated prescriptions demonstrate that RxVLM achieves a test accuracy of 91.43%, outperforming conventional convolutional and transformer baselines. These findings underscore the potential of VLM-inspired architectures to enhance the precision, safety, and efficiency of prescription digitalization in practical healthcare environments.</p>