This study proposes an efficient collaborative vocational education model based on AI algorithms, focusing on the deep integration of curriculum design and employment adaptability. Methodologically, the intelligent organization and structured association of course content is achieved through knowledge graph technology, and a joint LSTM-BERT-Attention extraction framework is constructed to solve the difficulties of text named entity recognition and relationship extraction. The model validation shows that the learning rate after hyperparameter optimization is 2×10-⁵, batch_size=16, and the F1 value of named entity recognition of LSTM-BERT-Attention on DCNE reaches 85.08%, which is significantly improved compared with the 78.73% of BiLSTM-CRF and the Flat model’s 83.67% is significantly improved. The F1 value of this paper’s model reaches 83.58% in course concept extraction for the dataset with 14 labels. In the knowledge graph construction, Neo4j visualization verifies the hierarchy and completeness of the knowledge network. The employment suitability experiment shows that the degree of modularization of 100%, the proportion of practical teaching of 20% and the intensity of technology tool integration of 1 correspond to the students’ employment suitability ability of 17.31±1.07, 17.07±1.27 and 17.57±1.13, respectively, and the AI-driven curriculum design significantly optimizes the matching between the teaching structure and job skills.