We introduce a novel graph neural network architecture where messages are bottlenecked mostly within a simplex itself. This allows messages to be passed in between nodes that are part of a simplex, increasing node classification probability. The assumption one needs to make is that the simplex itself has a semantic meaning. The main file may be found here. As for the preprocessing step, we encode a functor that unrolls a simplicial set back into a graph. After applying the Adj functor to a directed graph, one produces a multipartite graph, on which we can message pass using the simplicial message passing network.