This work presents a non-invasive, multimodal deep learning method for blood group prediction from images of agglutinated blood, fingerprints, and palmprints. Utilizing CNNs, the ResNet50 architecture, the system extracts important features across modalities to enhance prediction accuracy. This approach minimizes invasive testing, providing a fast, contactless solution with implications in emergency diagnostics and remote healthcare. Experimental results confirm the model's efficacy, demonstrating its potential for future AI-based biomedical innovations.