Multimodal systems require robust data representations capable of integrating heterogeneous sensory inputs while preserving relevant correlations and adaptive interpretability. This work proposes a novel self-structured data representation model inspired by biological cognitive processes and self-organizing maps (SOMs). The model uses adaptive-competitive dynamics to autonomously learn and store patterns from multiple modalities, dynamically updating these representations based on input similarity. A multimodal activation buffer quantifies concurrent activations to capture cross-modal relationships, enabling response composition that reflects learned correlations. Validation through synthetic multimodal data demonstrates the model’s ability to autonomously generate coherent pattern representations and response vectors, highlighting its potential for adaptability and autonomy in uncertain environments. This approach contributes to advancing interpretable, biologically inspired architectures for multimodal integration and autonomous systems.