A template generation system for generating document templates from a mixed set of document types including a template generation server programmed to receive a batch of documents, identify a plurality of text blocks, generate a plurality of clusters, generate a plurality of document arrays corresponding to the plurality of clusters, and compare each document array to each other document array to determine a percentage match. When the percentage match between two or more frameworks exceeds a threshold, the template generation system defines a subset of documents, and for each subset of documents, template generation system generates a template for the subset of documents. The template is a collection of the text blocks that are commonly included in each of the documents of the subset.