The Hascheck software, also known as Ispravi.me, performs very well in detecting and correcting "non-word" spelling errors. However, for "real-word" spelling errors, Hascheck rarely identifies and corrects the mistake. The aim of this work was to develop software that would enable the resolution of "real-word" spelling errors using n-gram based system. The developed software solution is deployed on a supercomputer maintained by the organization Srce. The n-gram system consists of a ScyllaDB database that contains n-grams and their frequencies. The application layer that communicates with the database is written in Rust, and a web server is built above the application layer to provide access to the end user. The system has predefined confusion sets, and upon request, it receives a text input. From this text input, it extracts the context wherever it finds a word from any confusion set. Using the context, all necessary n-grams are created for each word from the confusion set within which the word was originally found. The frequencies of the constructed n-grams are obtained from the database and these frequencies are analyzed. The system allows for modular creation of algorithms that will derive conclusions from the n-gram frequencies. Adding and changing algorithms is straightforward. The system implements three algorithms for prediction calculation, each using a different mathematical calculation method. The software solution can operate with many confusion sets, but a small subset was used during testing. The same text input was provided to each of the three defined algorithms. The solutions exhibited similar behavior across all three algorithms. The results show that the system is accurate in 70 % of cases. The system makes errors when users have repeatedly entered a misspelled word for a specific context. If the misspelled n-grams were removed from the database, the system would certainly function much better.