India\'s 600+ million social media users face unprecedented threats from sophisticated fake profiles and coordinated botnets that undermine platform integrity, spread disinformation, and influence elections. Traditional machine learning approaches relying on isolated account features fail to capture complex relational patterns and coordinated behaviors characteristic of modern botnets. This research proposes a novel Graph Neural Network (GNN) framework that models social networks as G=(V,E) graphs, where nodes represent user profiles with rich behavioral features and weighted edges capture interaction patterns. The architecture combines Graph Convolutional Networks (GCN) for neighborhood aggregation with Graph Attention Networks (GAT) for dynamic relationship weighting, enabling hierarchical feature learning across three GNN layers. Trained on combined TwiBot-22, Cresci-2015, and India-specific datasets, the model achieves state-of-the-art performance: 96.3% accuracy, 95.7% precision, 96.8% recall, and 96.2% F1-score, outperforming SVM (82.1%), Random Forest (85.3%), and other baselines by 11-18%. Key innovations include multi-scale graph embeddings capturing both individual account anomalies and bot cluster topologies, temporal interaction modeling, and real-time deployment as a scalable web application (<500ms inference/profile). Feature importance analysis reveals follower-following ratios, clustering coefficients, and posting variance as strongest discriminators. Successfully detecting a 47-account botnet with 95.7% recall, the framework addresses India\'s unique multilingual, high-density social ecosystem challenges. This GNN-based solution provides social media platforms with production-ready tools for maintaining authenticity, combating misinformation, and ensuring digital trust at national scale.