Learning graph representations of n-ary relational data has a number of real world applications like anti-money laundering, fraud detection, risk assessment etc. Graph Neural Networks have been shown to be effective in predicting links with few or no node features. While a number of datasets exist for link prediction, their features are considerably different from real world applications. Temporal information on entities and relations are often unavailable. We introduce a new dataset with 10 subgraphs, 20912 nodes, 67564 links, 70 attributes and 9 relation types. We also present novel improvements to graph models to adapt them for industry scale applications.