Optical Character Recognition (OCR) technology transforms how we interact with textual data by enabling machines to interpret text from images, scanned documents, and handwritten notes, with applications ranging from document digitization to real-time translation in augmented reality. This text explores building a custom OCR model to recognize the Wingdings font, a symbolic font developed by Microsoft, using the Vision Transformer for Scene Text Recognition (ViTSTR) architecture. The custom OCR model is particularly valuable in niche applications where traditional models fall short, such as translating symbolic text into readable English for accessibility or design purposes. While vision-language models like Flamingo excel at processing images and text, custom OCR remains essential for accuracy in specific languages, resource-constrained environments, data privacy, and cost-effectiveness. The process involves creating a Wingdings dataset from scratch, preprocessing images, and fine-tuning a Vision Encoder-Decoder model for text recognition tasks, with a focus on balancing accuracy and efficiency. The project demonstrates the adaptability of OCR systems in specialized use cases and highlights the potential for further exploration with different model architectures to optimize performance.