LARQL treats neural network weights as a graph database to enable complex structural queries. This tool allows researchers to isolate specific weight patterns without manual tensor slicing. It simplifies the analysis of model architecture and weight distribution. Practitioners can now audit internal model states using a familiar query language instead of raw Python scripts.