Embedding Agentic AI in Web Applications for Student Learning (CPD Case Study)
Dr Clément Godbarge (School of Modern Languages)
- What motivated you to use AI in this module, and what goals or challenges were you aiming to address?
Module ML5850 is part of an online postgraduate taught programme. Our students are distributed globally (Australia, China, Emirates, United States, sub-Saharan Africa, and elsewhere), and I cannot always be available to answer their questions in real time. This can be problematic when the learning process requires close assistance from the teacher. In the case of this module, teaching a query language known as XPath to students who have no previous experience with computer languages can be particularly challenging. Students can become discouraged by error messages and need guidance to help them understand their mistakes and persevere in writing correct queries.
- How did you design or adapt the assessment, and how did you prepare students for using AI appropriately?
The key to my approach is integrating the AI within a rigid software structure. This is not a chatbot but rather a form of agentic AI embedded within a web application. The AI follows very clear and stringent instructions. Its output is limited to providing short responses whenever the XPath query a student provides yields an error message. It is designed to examine the answer and the error, then provide feedback that does not disclose the solution but rather invites the student to try again whilst paying attention to what might have gone wrong. Students do not require any previous training to use this AI. Indeed, they are unaware of its presence, and they do not need to know it exists.
- What challenges did you encounter, and how did you address them?
Because it is a web application accessible to students worldwide, I have implemented strict safety measures, such as requiring the use of the University’s VPN, and ensured the application is resistant to various attacks, including prompt injection. Maintenance requires some time and regular updates. I spent considerable effort seeking assistance from the administration to achieve this experimentation. It was not straightforward. I waited more than six months to obtain what I needed from IT (server access, security settings, domain, etc.). Furthermore, I am currently funding the API costs from my own pocket.
- What benefits did you see for students and for your own teaching practice?
The benefits for online programmes are evident. These applications facilitate self-study through trial and error and make it easier for students to gain confidence in their work. They do not have to wait for my weekly virtual office hours to ask about the exercises; the AI provides useful feedback instantaneously. This is also beneficial for me, as I receive fewer emails during weekends or in the middle of the night.
- How did you evaluate the usefulness of this assessment to ensure that it reflected the desirable learning outcomes?
I have measured its effectiveness in an informal manner, as we do not have enough students to create a statistically significant survey. Students appreciate this application and have expressed interest in seeing more of them integrated into our programme.
- What would you do differently next time, and what advice would you give to colleagues?
If you are not a software developer, I would not recommend attempting this. My attempt was experimental, and this experimental nature is the only reason I found the energy and goodwill to create this in my spare time. Ideally, the University should provide infrastructure to experiment with our own models that we run locally—it would be more cost-effective and secure. The University should also create a ‘sandboxing programme’ that helps academic personnel willing to experiment obtain the necessary authorisations quickly, perhaps for just a semester or two. If the experimentation proves successful, the necessary paperwork could then be completed to make it compliant.