We need a set of core machine learning nodes, e.g.:

• to cluster embeddings / documents (k-NN, DBSCAN)
• to classify images in combination with image embeddings (SVMs)
• for dimensionality reduction to visualize high-dim embeddings (t-SNE, PCA)
• for regression
• for preprocessing (normalization)
• for type conversions to handle vectors and arrays / batch predictions

ML-nodes should support run in train and eval mode with option to persist trained checkpoints.