Techniques are described for predicting test scenarios and generating test case documents and/or automation scripts using machine-learning algorithms. For example, input may be received representing a web page, and an HTML Document Object Model (DOM) of the web page may be analyzed. From the DOM, a plurality of HTML elements may be extracted and processed by a machine-learning algorithm. Based on the processed plurality of HTML elements, a plurality of predictions for test case scenarios may be generated, and converted into a set of human-readable instructions, such as a test case document, and/or a set of machine-readable instructions, such as an automation script. In some instances, a user selection of at least one predicted test scenario from a displayed list of one or more predicted test scenarios is received and corresponding human-readable instructions and/or machine-readable instructions are generated for the selected scenario(s).