tesseract-ocr/tesseract ? reverse-engineered prompt

Build me a command-line tool for Optical Character Recognition (OCR) based on the open-source Tesseract engine. I need something that can take an image file, like a PNG or JPEG, and extract all the text from it. It should use a modern machine learning model, like the LSTM one I've heard Tesseract uses, to get the best accuracy.

The most important features are broad language support—it should handle over 100 languages right away—and flexible output. I want to be able to save the extracted text as a plain text file, but also as a searchable PDF. It doesn't need a graphical interface, just a solid command-line program. This seems like a pretty complex C++ project, so please look up the current documentation for the best way to build it from the source files.