The PatentAI technology uses custom natural language processing and machine learning to detect Intellectual Property (IP) infringement. Our experiments and demonstrations shows a significant improvement in the performance of the system by using a custom natural language pre-processing step that converts the patent (legal) language into simple English (which is a desirable function by itself), simplifies the text to capture its essence, and transforms it into a concise graph representation. After this critical pre-processing step, we use a learning model trained on a Paraphrase Identification dataset to detect if two given patent excerpts paraphrase each other, and therefore, the corresponding patents infringe on each other. A key novelty of our approach lies in the techniques used in the pre-processing step to increase the performance of the learning model. In addition, we address the difficult problem of IP infringement detection by converting it to a paraphrase identification problem, and leveraging existing models and datasets. In our demonstration, the user can enter two excerpts from different patents to detect if one patent infringe on the other. The system processes the two texts and interactively shows the user the following: (a) results of converting the patent excerpts from a patent/legal language to simple English, (b) a graph representation of each text to capture the essence of the ideas conveyed in the text (c) key extracted features, and (d) the IP infringement prediction (Match or Mismatch), where the “Match” prediction means that the model predicts that one of the patents infringe on the other.