Brain-computer interfaces (BCIs) can decode not only what users are thinking but also the intensity of their cognitive effort. However, BCIs have traditionally been constrained to single-task applications. Multitask learning (MTL) offers a promising solution by enabling BCIs to handle multiple related tasks simultaneously, enhancing both performance and usability.This study applied MTL to EEG data from N-back working memory tasks (0-back, 2-back, and 3-back) using openaccess data from 26 participants at Technische Universität Berlin. We developed a novel hybrid CNN-LSTM-tAPEformer architecture that integrates Convolutional Neural Networks for spatial feature extraction, Long Short-Term Memory networks for temporal sequence modeling, and Transformer blocks with specialized attention mechanisms for capturing long-range temporal dependencies. The proposed model performs dual functions by classifying accurate behavioral responses while simultaneously measuring cognitive workload across varying task complexity levels. Notable innovations include the development of Time Absolute Position Encoding (tAPE) that enhances temporal processing by integrating sinusoidal positional encoding with adaptive channel-specific encoding to preserve temporal relationships in EEG data. The system incorporates regional and temporal self-attention mechanisms along with global attention pooling to achieve enhanced neural pattern detection. Through leave-one-subject-out cross-validation methodology, the model was trained using data from all participants except one, then evaluated on the excluded individual to assess cross-subject generalization performance. Findings validate the hybrid CNNLSTM-tAPEformer model's efficacy for practical multi-task learning implementations, establishing its utility for BCI applications that demand concurrent cognitive state identification and workload assessment.