GraNNite is a novel framework that enables efficient execution of Graph Neural Networks on edge devices with limited computing resources, bridging the gap between complex graph processing and resource-constrained neural processing units.

• Optimizes GNN performance through hardware-aware mapping techniques
• Enables on-device graph processing for privacy-sensitive applications
• Supports Retrieval-Augmented Generation for LLMs without cloud dependence
• Addresses key challenges of irregular memory access and dynamic graph structures

This research is significant for Engineering as it allows deploying sophisticated graph-based AI capabilities on edge devices like laptops and client PCs, enabling real-time processing without compromising performance or privacy.

GraNNite: Enabling High-Performance Execution of Graph Neural Networks on Resource-Constrained Neural Processing Units