An early exploration (high-school era) of OCR applied specifically to source code — images of code that need to become editable text while preserving indentation and special characters that generic OCR tools mangle.
The goal was to handle the failure modes that make code OCR hard: preserving whitespace structure, disambiguating similar glyphs (0 vs O, 1 vs l vs I), and passing curly braces and punctuation through untouched.