Image Translator

I once tried Google’s and Yandex’s image translation features and thought, “I could build that.” The idea was to use OCR to detect text, extract it, and store the bounding box positions. Then I’d translate the extracted text and render the translation back into the image at the original locations. Turns out it required far more computing power than I expected. Each step took around 10–15 seconds, which felt way too slow. On top of that, the codebase became a mess because I built most of it from scratch—so after a while, I gave up.