A semantic sorting approach based on embeddings, a distance matrix between entities (cosine distances) and a solver, originally developed for solving vehicle routing problems (VRP) like the Traveling Salesman Problem (TSP). The matrix is visualized as table and the embeddings are reduced in their dimensionality via UMAP to plot them on a 2D plane. Just paste a new-line separated list of words or terms into the text field and optimize. Results are usually quite good but not perfect as it is yet lacking a "tree-based" approach where large topic clusters are identified first. I had this idea for a while and simply needed to get it out of my system. If you'd like to contribute, I'm happy to merge creative PRs! The tech stack really sums up my geospatial and NLP interests: transformers.js, onnx, umap, vrp, deck.gl. Developed by Dominik Weckmüller. Source code available on GitHub.