UG06 – AI Grader

Automatic grading tools are widely adopted in educational institutions, and various studies have proven that they can reduce the workload of school staff and improve students’ learning experiences. Existing automatic grading tools, such as nbgrader and Gradescope, can achieve automatic grading of programming-related queries, but feedback must be created manually. Limited by the needs of large-class teaching environments, a large number of manual tasks can also affect the quality of feedback, making it difficult for students to receive targeted and useful insights. This also increases the difficulty for teachers in summarizing students’ common mistakes. This project aims to grade students’ responses and provide comprehensive feedback. It integrates a Large Language Model (LLM) into nbgrader for grading, aiming to significantly enhance the feedback mechanism. To enrich the feedback with more detailed and useful insights, the system will leverage Retrieval Augmented Generation (RAG), which will fine-tune the feedback to be deeply informative by creating a knowledge base to provide constructive suggestions, thereby improving the efficiency and effectiveness of the grading process. Additionally, students’ answers and grading history will be recorded in a database to automatically analyze common errors and provide students with a chatbot for more personalized help.