Crossword puzzles pose a challenging problem in vocabulary and logic, and automatic generation of these puzzles is complex due to their combinatorial nature. This paper will explore their compilation by representing it as a Constraint Satisfaction Problem (CSP), as well as novel methods of generating crossword dictionaries, layouts and controlling puzzle difficulty. We will present a model for representing crossword generation as a CSP, including preprocessing steps, dictionary organization, and constraint modeling. Various optimization techniques, including heuristics, constraint propagation, and parallel processing are discussed. Furthermore, this work is the first of its kind for the Macedonian language, utilizing NLP and existing databases to determine the frequency of various words and control difficulty. Finally, we will use experimental tests and comparisons to existing puzzles to demonstrate the effectiveness of these techniques in generating crossword puzzles of varying sizes.