AI Tutor Hub: A provincial government-funded pilot of AI tutoring for anatomy and physiology

The problem

Anatomy and physiology courses are foundational for students in healthcare programmes such as nursing, paramedic, personal support worker, practical nursing, and occupational therapist assistant/physiotherapist assistant. Students need this knowledge for later study and for their future work in healthcare settings.

These courses are also among the most difficult early requirements for healthcare students. At Lambton College, anatomy and physiology is a high-failure-rate course for first-year students. When students fail or withdraw, it can delay graduation, reduce programme retention, and slow progression into health professions at a time when Ontario is already facing healthcare workforce shortages. These shortages have cascading effects through the healthcare system. Every student who fails or withdraws from a foundational course represents a delayed or lost healthcare professional at a time when the system cannot afford either.

The subject is challenging because students need to understand both anatomy, the structure of the body, and physiology, how body systems function. While anatomy can sometimes be approached through memorisation, physiology is highly interconnected, making memorisation alone insufficient. As a result, even otherwise strong students can struggle to pass.

Students who fall behind may need more individualised support than standard course materials can provide. They need help identifying which concepts they understand, which ones they are struggling with, and how different body systems connect. However, providing this kind of personalised, on-demand support at scale is difficult for faculty, especially in high-enrolment first-year courses.

The solution

Lambton College is a publicly funded college in Ontario. After early anecdotal evidence that pairing AI tutoring with visual learning materials helped students who had struggled in anatomy and physiology, the college applied for funding to test the approach more systematically. It secured a grant from the Higher Education Quality Council of Ontario (HEQCO), an arm's-length body of the provincial government that funds research on higher education, through its Consortium on Generative Artificial Intelligence. The project was one of six accepted from 41 proposals.

With this funding, the college created an AI Tutor Hub for anatomy and physiology courses, combining course-specific AI tutoring with curated learning materials including videos, visual references, and 3D anatomy resources. The project was built entirely on publicly funded infrastructure. The AI tutoring tool, AI Tutor Pro, is provided by Contact North, another arm's-length organisation of the Ontario provincial government, established to support access to education across the province. AI Tutor Pro is not a commercial product. It does not sell advertising or data and does not require students to create an account or log in, which removes a significant barrier to deployment given student data privacy concerns.

AI Tutor Pro enables faculty or project teams to create student-facing AI tutors that are constrained to specific material, rather than drawing on general knowledge. This was important because general AI tools can provide information that is technically accurate but not relevant to a particular student's course, assessment, or scope of practice. For students already struggling with the material, not knowing what is and is not relevant to their programme can make things harder, not easier.

To build the tutors, subject matter experts reviewed previous versions of the anatomy and physiology courses and produced weekly notes outlining what each group of students needed to know, taking into account the different scopes of practice across nursing, practical nursing, paramedic, personal support worker, and occupational therapy assistant programmes. The team initially planned on creating one AI tutor per course, but testing revealed that the tutor was not effective across such a broad range of content. Instead, they created a separate AI tutor instance for each week, aligned to the body system being studied, which kept the tutor reliably within scope.

The tutor is designed as a Socratic tool. It does not simply provide answers but helps students test their understanding, identify gaps in their knowledge, and work through difficult concepts. Students enter the topic they want to review, choose a starting level, and work through questions with the tutor at their own pace.

The hub was embedded directly into students' learning management system course shells, so students did not need to visit a separate site or create another login. Before gaining access, students completed a short introductory module explaining what the tool was, how to use it effectively, and relevant privacy information. Each weekly section focuses on a body system, such as the cardiovascular system, and includes the relevant AI tutor instance, along with supplementary materials.

These supplementary materials were an important part of the design. The team therefore focused on resources that could be embedded directly into the course shell, including publicly available videos and web-based 3D anatomy models. Subject matter experts and a project coordinator identified appropriate materials, which an e-learning production assistant then built into the weekly modules. This curation work ensured that students had both the AI tutor and the visual references they needed in one place, without having to search for resources themselves.

The impact

1. Early evidence of improved student outcomes

The study is evaluating whether access to AI Tutor Pro improves pass rates, grades, course withdrawal rates, and the number of students failing anatomy and physiology. The research team is comparing the current cohort, which had access to the AI Tutor Hub, with cohorts from the previous three years that did not. They also separated failure rates from course withdrawals, because students may withdraw before receiving a final grade, which can mask the true failure rate if only pass/fail figures are examined. Initial results show a statistically significant improvement in student achievement for the AI tutor cohort vs. the historical cohort.

2. Strong student satisfaction and perceived usefulness

Qualitative feedback was positive, with positive responses in the 80% to 90% range across the main survey questions. Students said the AI Tutor Hub was easy to use, valued its availability whenever they needed it, and felt it helped them understand the course material. Confidence improved less than other measures, which is consistent with focus group feedback. The comparatively lower increase in confidence could be a reflection that anatomy and physiology, as subjects, remained challenging even with additional support.

3. Students valued that the tutor was course-aligned and trustworthy

A key theme from the surveys and focus groups was trust. Students valued that the tutor was aligned to their course content and based on materials that had been reviewed and scoped by subject matter experts. Students also reported that the tool felt trustworthy because it was embedded in their course environment and clearly scoped, rather than being a general-purpose tool they had to navigate on their own. This helped distinguish it from using a general AI chatbot, where students might not know whether the information was relevant to their programme, assessment level, or scope of practice. Some students wanted even closer alignment to learning objectives or tests, but the project team deliberately focused on giving students a broad grounding in the content rather than turning the tool into test preparation.

4. Student experience was assessed through surveys and focus groups

Alongside the quantitative analysis, Lambton College gathered qualitative feedback from students through surveys and focus groups. The surveys captured students' perceptions, satisfaction, ease of use, and whether the tool helped them understand the material. Focus groups asked similar questions but allowed students to explain their experiences in more detail, including what they found useful, where the tool could improve, and how it affected their confidence.

Lessons learned

• One AI tutor per course was not granular enough. The team initially expected to create one tutor per course, but found that the AI could not stay reliably within scope across an entire course's content. Creating a separate tutor for each week, aligned to a specific body system, kept the tutor focused and gave students more reliable responses.

• Supplementary visual material made a meaningful difference. Pairing the AI tutor with curated videos and 3D anatomy models was important, particularly for content with a strong visual component. The combination of being able to ask questions and see what was being described was more effective than either alone.

• Faculty communication and buy-in were essential. Getting faculty on board early and embedding the tool directly in the course shell, rather than hosting it separately, made a significant difference. Students were more likely to use the tool and see value in it when their faculty member had already endorsed it and explained how it fit into their learning.

• Model updates can break what was working. When the underlying ChatGPT model was updated, it disrupted the behaviour of the AI tutor. The new model was technically better, but it initially did not interact with the software layer and prompts in the same way. Ongoing monitoring of outputs is essential because changes to the underlying model can have unexpected effects on how well a tutor performs.

• Students need to be taught how to use AI tools effectively. The introductory module was important. Students did not intuitively understand that they should start a new chat when switching topics, or that the AI reads the entire conversation history when generating each response. Without this guidance, students found the tool confusing when conversations became long or shifted between subjects.

• No plan survives first contact with users. Testing with real students surfaced issues the team had not anticipated, because the team's familiarity with the tool meant they did not make the same mistakes or hold the same assumptions as first-time users. Regular check-ins with students and faculty throughout the pilot were essential for catching and fixing problems early.

• Using a publicly funded, non-commercial tool simplified deployment. AI Tutor Pro, provided by Contact North, is not a commercial product and does not require student logins or collect user data. This removed significant friction around data privacy and made it easier to deploy the tool at scale within a publicly funded institution.

Student data privacy was a key consideration throughout. AI Tutor Pro, provided by Contact North, does not require students to log in and does not collect or store user data. This was important for a publicly funded institution deploying an AI tool to students, as it avoided the need for additional data-sharing agreements or privacy impact assessments. The introductory module included information about privacy so students understood how the tool handled their interactions.