Diabetes management and detection often rely on sophisticated and costly diagnostic tools, which can be cumbersome and time-consuming. This work suggests a novel method to improve diabetes diagnosis by combining a triangular hollow-core photonic crystal biosensor with Terahertz re-fractive index measurements and advanced Capsule Genghis Khan Sparse Graph Attention Network (ThzBiosen-Cap2G2SA-Net) techniques. The biosensor presented in this research leverages the distinctive characteristics of Terahertz waves to detect changes in the refractive index of biological tissues, with a focus on identifying diabetic conditions. By integrating Terahertz refraction index information with the ThzBiosen-Cap2G2SA-Net model, the study aims to increase the biosensor's efficacy and enable accurate and effective diabetes detection. The biosensor's data is processed by the ThzBiosen-Cap2G2SA-Net model, which accurately distinguishes between tissues with and without diabetes. This integration significantly improves sensitivity and specificity in detecting diabetic markers, resulting in superior diagnostic accuracy compared to traditional methods. Through comprehensive evaluations, the study demonstrates that the proposed system significantly outperforms traditional biosensor configurations, achieving a sensitivity of approximately 98% in detecting diabetes. This method shows how cutting-edge photonic technology and machine learning techniques can be combined to transform diabetes diagnosis and provide a quick, accurate, and affordable diagnostic solution.