Automated Educational Quiz System Using RAG for Student Assessment

Abstract

Searching and finding relevant information through text books and other sources has always been a hassle for students. Ironically, text books are one of the best sources for students to find appropriate information. We wanted to build a system that could fetch the necessary data from text sources precisely and also help students to test what they have learnt. So, DocVQA is one way in which we could implement this. Document visible question answering (DocVQA) pipelines, which reply to enquiries derived from documents, has many makes use of. Current methodologies concentrate on handling single-web page documents the usage of multi-modal language fashions (MLMs) or rely on text-based totally retrieval-augmented technology (RAG) that use text extraction technologies like optical character reputation (OCR). To optimise our pipeline's performance, we used M3DOCRAG. M3DOCRAG is an modern multi-modal RAG framework that adeptly handles diverse report contexts (both closed-area and open-area), query hops (single-hop and multi-hop), and proof modalities (textual content, charts, figures, etc.). M3DOCRAG identifies relevant documents and responds to enquiries via a multi-modal retriever and a masked language model, allowing it to efficaciously manage single or many files even as retaining visual records. This device facilitates the availability of a QA assistant and an MCQ generator, enabling college students to assess their information. Students can be able to do examinations (both more than one-choice and descriptive) the use of our method. Descriptive responses are assessed the use of a textual content similarity degree that quantifies the resemblance among human-written and AI-generated replies, ultimately assigning rankings based in this evaluation.