EduMark AI: Transforming Assessment and Efficiency Through AI-Enhanced Learning and Personalised Student Feedback in STEM Education

This project is funded by the Drapers' Fund for Innovation in Learning and Teaching

Responding to a need

Assessment and feedback are integral to student learning, but traditional marking in higher education is highly time-intensive and often inconsistent. Within SEMS alone, grading consumes approximately 3,400 staff hours annually, about 67 hours per educator for a class of 50 students. Feedback quality can vary between markers, leading to student dissatisfaction and difficulty in tracking progress. Large cohorts amplify these challenges, limiting opportunities for timely, detailed, and personalised feedback.

EduMark AI was conceived to address these systemic issues. The project directly responds to institutional priorities outlined in Strategy 2030 and the Active Curriculum for Excellence (ACE), particularly in relation to inclusivity, engagement, and efficiency. It aims to provide students with timely, consistent, rubric-aligned, and actionable feedback, while also alleviating workload pressures on staff.

By integrating AI into assessment in a hybrid model, the project leverages technology to support, not replace, academic judgment. Educators retain oversight, ensuring AI feedback is aligned with learning outcomes and pedagogical integrity. Simultaneously, students benefit from faster, clearer, and more constructive feedback, fostering reflective learning and greater agency. The project also provides a structured methodology for staff, addressing barriers of confidence and familiarity with AI in education.

The Approach

EduMark AI was piloted across the School of Engineering and Materials Science, London (SEMS), and the Queen Mary Engineering School, China (QMES), engaging over 200 students and multiple faculty collaborators. Assessments included lab reports, coursework essays, problem-solving tasks, and research posters.

Design

The approach followed a systematic, three-phase methodology:

Phase 1: LLM Selection and Evaluation Comparative analysis of AI systems (ChatGPT, Google Gemini, Graide, Timely Grader) against human marking benchmarks. ChatGPT Plus emerged as the optimal choice due to its superior performance in generating assessment-aligned feedback and strong adaptability across diverse assignment types. Equally important, ChatGPT was prioritised for its robust data privacy and cybersecurity compliance. To safeguard student information, LLM settings were enabled to ensure that all conversations remain confidential and are not used for model training.

Phase 2: Prompt Engineering Framework: Development of structured templates aligned to marking rubrics, integration of exemplar answers, and a systematic feedback format (“strengths,” “areas for improvement,” “action points,” and “grade justification”).

Phase 3: Hybrid Human-AI Implementation: Rather than full automation, we implemented a hybrid model that maintains educator control and preserves academic integrity. The system includes an "Educator Review" stage, where AI-generated feedback can be validated, amended, or enriched before being released to students. This preserves academic integrity while capturing efficiency gains.

Co-creation and Student Partnership: Students were genuine co-creators throughout the development process. They contributed to surveys and coded the EduMark AI App interface. Their iterative feedback shaped the system to ensure it was student-friendly, actionable, and aligned to learner needs.

Implementation

The pilot ran as a “shadow marking” exercise, where AI and educators simultaneously assessed anonymised submissions. Student surveys evaluated the clarity, fairness, and usefulness of feedback. Educators assessed time savings and grading consistency.

Resources: The project used existing AI platforms, survey tools, and a student-developed web app. Ethical approval (QMUL Ethics of Research Committee) ensured compliance with data privacy and transparency standards.

Alignment with Strategy 2030 & ACE: EduMark AI directly supports Strategy 2030 goals of educational innovation and embedding AI literacy. It enhances ACE by enabling timely, inclusive, and personalised feedback, fostering student engagement and reflective learning.

Impact

EduMark AI demonstrated a transformative impact for both students and staff.

1. Grading Consistency: AI grading aligned within 0–10% variance of educator grades, providing greater reliability across large cohorts.
2. Efficiency Gains: Marking time reduced by 50–60%. For example, 40 lab reports took ~230 minutes using AI, versus ~610 minutes manually.
3. Educator Benefits and Feedback: As a result of the time savings arising from the use of EduMark, staff could focus on enhancing teaching, learning and student engagement through mentoring, curriculum development and   innovation. Staff also reported reduced marking fatigue and increased confidence in the consistency of feedback. Some educators described EduMark AI as a “game-changer for educators” and said they were “blown away by the detail with which it provided feedback.”
4. Educational Enhancement: The system generates immediate, detailed, personalised feedback, allowing students to understand their performance while assignments remain fresh in their minds. This supports reflective learning and enables the timely application of insights to future work, aligning with Queen Mary's Active Curriculum for Excellence approach.
5. Student Satisfaction: Student feedback consistently praised the AI's specificity and actionable suggestions. Students particularly valued receiving "concrete examples and suggested corrections" that helped them "identify what went well and what needed improving." The quality of feedback  often surpassed expectations, with 86% of students rating the AI feedback as good or excellent in terms of usefulness, quality, and relevance.
6. Broader Institutional Impact: EduMark AI addresses systemic challenges in higher education while demonstrating ethical AI adoption. The efficiency gains allow educators to focus on higher-value activities such as curriculum development, personalised mentoring, and innovative teaching approaches.
7. Strategic Impact: The project has positioned QMUL as a national leader in AI in education. It has been showcased at the Queen Mary Festival of Education, accepted for publication on the Jisc and Advanced HE blogs, and selected by OpenAI Academy’s international “Professors Teaching with AI” series.

These outcomes show EduMark AI’s potential for broader adoption across disciplines, supporting Queen Mary's ambition to embed AI literacy and drive excellence in education.

Recommendations

Key Success Factors

• Ensure ethical safeguards by maintaining human oversight of AI-generated outputs
• Start with pilots across diverse modules to build evidence and staff confidence
• Co-create with students to ensure usability and educational value
• Provide staff training in prompt engineering and AI literacy
• Scale gradually, with clear evaluation frameworks for consistency, fairness, and inclusivity

Implementation Considerations

• Ethics approval essential: Ensure formal institutional ethics committee approval for student data use
• Discipline-specific adaptation: Results may vary across subjects—thorough testing in your context is crucial
• Continuous refinement: Plan for iterative improvement based on student and educator feedback
• Training and support: Provide comprehensive training materials for educators' confidence with AI tools