Overview
RAGrade was my final year project, developed in 2026 to simplify and improve the consistency of marking open-ended exam responses. The system allows teachers to upload exams, create marking rubrics, and submit student answers. RAGrade then evaluates each response, assigns a score, and provides a written explanation for the grade.
To improve grading quality over time, the system uses Retrieval-Augmented Generation (RAG) by retrieving relevant previously graded examples from a vector database and using them as reference during evaluation. It also includes a trust-scoring mechanism that prioritizes teacher-validated grades over AI-generated outputs, allowing feedback and corrections from teachers to gradually improve future assessments.
The project was tested using IB and GCSE-style short-answer papers, where wording and partial understanding often affect grading decisions. Evaluation results showed up to 99% agreement with human markers, demonstrating improved consistency while reducing manual marking effort.
Tech stack
Gallery
