Abstract. This research presents an intelligent system for automated generation of domain-specific learning assignments in civil aviation education using fine-tuned T5-small transformer models. Traditional assignment creation requires significant instructor time and expertise in aviation regulations, technical specifications, and safety procedures. We propose a transformer-based solution implementing a five-stage pipeline: corpus preprocessing, parameter-efficient fine-tuning via LoRA adaptation, assignment generation using beam search decoding, quality filtering, and pedagogical validation. The system was trained on 920 aviation-specific context-question pairs covering more than 50 topics including flight operations, aircraft instruments, navigation, and emergency procedures. Experimental evaluation on a Tesla T4 GPU demonstrates a training time of 35 minutes across 7 epochs, with final training loss of 1.3506 and validation loss of 1.221. Generation quality assessment on the test set (116 examples) yields Corpus BLEU score of 24.27, ROUGE-L F1 score of 0.5087, and BERTScore F1 of 0.6017. Aviation terminology coverage analysis shows that 38.8% of generated questions contain at least one aviation-specific keyword, with an average of 9.4% unique aviation terms per question. Additional metrics include a unique bigram ratio of 0.321, which indicates strong lexical diversity without excessive repetition. Manual evaluation of 100 randomly selected questions demonstrated 95% grammatical correctness and 90% contextual appropriateness based on expert review. Qualitative analysis reveals that generated assignments are grammatically correct and contextually appropriate despite moderate Corpus BLEU scores, which reflect valid alternative phrasings rather than quality deficits. Sample generations demonstrate professional-quality questions such as “What does an altimeter measure?” and “What happens when a stall occurs?” The system reduces instructor workload in assignment creation while maintaining technical accuracy and domain relevance, providing a foundation for AI-assisted educational content generation in specialized technical domains.

Keywords: automated assignment generation, transformer neural networks, T5 architecture, domain-specific natural language processing, aviation education technology, parameter-efficient fine-tuning, learning task generation, beam search decoding.