Real-Time EEG-Driven Neuro-Adaptive Learning at the Edge: A Dual-Loop CNN–LSTM Architecture for Cognitive Workload Regulation in Higher Education

Background: EEG-driven instructional interfaces can complement behavioral adaptation with workload-related information. Methods: The reported study involved 34 engineering students in neuro-adaptive and static-interface conditions, using 14-channel EEG at 256 Hz and 2 s windows advanced every 0.5 s. This revision specifies structured task/effort/performance annotation and an executable Symlet-4 preprocessing procedure. Results: A supplied 169,795-parameter CNN–LSTM checkpoint was executed over 1000 continuous windows identified by the authors as a 501.5 s extract from one participant. Mean processing time was 3.945±0.633 ms, with 4.843 ms P95 on the reported AMD EPYC CPU using one PyTorch thread; no measured total exceeded 500 ms. This execution does not reproduce the original classification, calibration or educational analyses. Conclusions: The supplied implementation supports a limited computational-feasibility finding. The original participant outcomes remain reported summaries whose annotation, processing and model provenance have not been independently reconstructed in this revision; broader efficacy and cross-participant generalization therefore remain unverified by the available materials.