I was searching for a way to gain experience with large language models (LLMs) and was influenced by the transition at MSOE from quarters to semesters (2023). This transition meant that, while teaching most of the same material, instructors needed to reorganize and reconfigure most of what we taught.

In an attempt to address both problems, I created a web-based interface, leveraging Gradio, and a Python backend with a database for maintaining content generated for multiple users for different courses. I use programmatically generated prompts for content generation, each based on what users put in for information about the course and subsequently created modules or lectures. I programmed the software so that users can enter their own information, use an LLM to automatically generate information based on previous entries, or edit LLM-generated content.

The images below show a simple progression in the software where I entered information about a course I teach, Fluid Mechanics. I then used the LLM functionality to generate a set of course modules and a lecture from selected topics. The generated lecture is used to populate a Google Slides document and images from Google Search are inserted into corresponding slides. Users can edit and download the slides OR they can be exported as a lecture video, leveraging text-to-speech generation for each slide transcript. Quizzes and Assessments can be generated based on a single lecture or specified Module topics and exported as a Canvas LMS quiz.

At the bottom of this page is an un-edited lecture generated using the Course Designer software. Module and Quiz/Assessment generation features proved to be very helpful during the transition, but generated lectures needed considerable editing - so much so that they were not used.

Current efforts are focusing on the implementation of recent RAG, Knowledge Graph, and Reflection techniques to improve on the longer form content generation needed for better lectures.

Languages used: Python