AI-Driven Automated Feedback and Tutoring System for Higher Education

Team

Supervisors

Tags

Final Year Project · Artificial Intelligence · Automated Grading · Moodle Integration · LLM · Plagiarism Detection · Education Technology

Table of content

  1. Abstract
  2. Related works
  3. Methodology
  4. Experiment Setup and Implementation
  5. Results and Analysis
  6. Conclusion
  7. Publications
  8. Links

Abstract

Large undergraduate classes make manual grading and feedback slow, inconsistent, and difficult to scale. This project proposes an AI-driven system integrated with Moodle to automate structured, rubric-aligned feedback and support student learning. The pipeline retrieves assignment submissions via Moodle REST APIs, performs deadline-aware processing, checks for potential plagiarism using vector similarity, and generates feedback through a modular multi-agent LLM workflow. The system also maintains concept-level learning history to produce adaptive and personalized guidance. The final outcome is an end-to-end platform that reduces instructor workload while providing faster, consistent, and actionable feedback to students.

Automated assessment systems traditionally rely on rule-based marking, keyword matching, or classical machine learning models, which often struggle with open-ended answers and explanation-based questions. Recent LLM-based approaches have shown improved capability in generating human-like feedback, but they may produce inconsistent scoring, hallucinations, and weak transparency without proper structure. Plagiarism detection tools commonly use string matching; however, paraphrased copying and concept-level similarity require semantic methods such as embeddings and vector similarity. Our work builds on these directions by combining deadline-aware automation, multi-agent feedback generation with QA, semantic plagiarism detection, and learning-history-based personalization within a Moodle-integrated workflow.

Methodology

Our methodology follows a fully automated pipeline:

  1. Moodle Integration
    Retrieve course, assignment, and submission data using Moodle REST web services with token-based authentication.

  2. Deadline-Aware Orchestration
    A scheduler periodically checks assignments and triggers processing based on the due date:
    • Before deadline: ingest + plagiarism check only (no grading)
    • After deadline: plagiarism check + multi-agent grading

  3. Submission Processing
    Extract/normalize text (online text or file-based submissions) and segment answers by questions (Q1/Q2/Q3…).

  4. Plagiarism Detection (Semantic)
    Generate embeddings per question and compare against other submissions using cosine similarity; flag suspicious cases and store evidence.

  5. Multi-Agent Grading + Feedback
    Use specialized agents (correctness, misconceptions, clarity, personalization, synthesis, QA) to produce rubric-aligned scores and structured feedback.

  6. Storage + Feedback Delivery
    Store all results in the database and return grades/feedback back to Moodle (and/or an internal dashboard).

Experiment Setup and Implementation

System Stack

Key Implemented Components

Results and Analysis

Current preliminary results (mid-progress)

Planned analysis (final phase)

Conclusion

This project demonstrates the feasibility of an AI-driven, Moodle-integrated automated feedback system that combines deadline-aware fairness, semantic plagiarism detection, and structured multi-agent grading. The remaining work focuses on robust evaluation with real anonymized submissions, optimizing prompt/agent behavior, and producing quantitative results to validate pedagogical and technical effectiveness.

Publications