Frustrated by the limitations of Tesseract OCR to extract text from meme images, the author found a way to leverage the iOS Vision API capabilities from older iphones models connected to a Raspberry Pi to build his own OCR service.