Fixed a critical bug where next-word predictions would overwrite the previously typed word.
Fixed CORS / Security Error by properly configuring web accessible resources.
Model dataset quality significantly improved with 100% natural, daily conversational datasets.
The AI bigram model was aggressively pruned to reduce memory footprint and increase prediction speed.

Dataset quality significantly improved: News and formal texts were removed and replaced with 100% natural, daily conversational datasets .
Performance optimization: The AI bigram model was pruned to remove rare words, decreasing the memory footprint and significantly speeding up autocomplete suggestions.
Enhanced prediction accuracy for Persian, Turkish, and English languages.

Fixed a critical bug preventing the offline AI model from loading in production. Added proper access rights, so the text predictor now works flawlessly on all websites.