A Client-Server Based Educational Chatbot for Academic Institutions

The use of Generative AI applications in academia, such as ChatGPT, Bard and Perplexity among others is growing at a rapid pace. With its rise, certain ramifications are being felt in regards to the quality and correctness of knowledge and output of these apps. This is detrimental to the process of learning and understanding subject matter in relevant context. In order to stop this issue in its tracks, a solution must be identified, built, tested and deployed so that students will get reliable outputs from trusted sources. This research analyses and describes a sample architecture as well as implementation for such a solution. It incorporates the latest in AI research and development, such as the Large Language Model Mixture of Experts (MoE) architecture as well as a Retrieval Augmented Generation (RAG) with a vector store to use trusted documents such as presentation files, PDF handouts and more from instructors. This is used to add context to the request to the Large Language Model and enrich the understanding as well as response of the model. Apart from this, the application is packaged into a server that can be run on the intranet, as well as deployed for public access. A frontend client page is served to the user, and communicates with the server for all its functioning.