Cairn: Open Infrastructure for adaptive AI Tutoring
Cairn turns AI into an auditable tutor that adapts support to each student’s prior knowledge—without giving away the answer.
Duration: April 2027 – December 2027
Education level: Tertiary level
Topics: AI – Artificial Intelligence; Digital Tools
Keywords: AI tutoring; personalised learning; adaptive scaffolding; open-source education; higher education
Background
Students enter higher education with very different levels of prior knowledge, but they are often given the same static learning materials and the same amount of support. What helps a novice can be redundant, or even counterproductive, for an advanced learner. Lecturers know this, yet adapting exercises and guidance for every student is rarely feasible at scale. Generative AI could make personalised support affordable, but unconstrained chatbots create a new problem: they can provide answers instead of helping learners think. Their teaching decisions are also difficult to inspect, change or audit, and many solutions depend on proprietary cloud services. Cairn addresses this gap with an open, auditable AI tutoring infrastructure. It separates pedagogical decisions from content generation: explicit rules determine how much support a learner receives, while a selectable language model generates the corresponding hint. Support fades as knowledge grows, and the learner always completes the final step. The system is designed for institutional control, different AI providers, including local models, and reuse across courses and learning platforms.
Objectives
Cairn will deliver an open-source AI tutor that adapts support to each learner’s prior knowledge without giving away the answer. Its pedagogical rules will remain visible, testable and independent of the chosen language model. The project will provide a web application, interfaces for learning platforms and AI systems, and reusable skills for diagnosis, scaffolding and feedback. A pilot in an introductory BFH mathematics course will test whether Cairn places learners appropriately, provides well-judged support and works reliably in practice.
Method
We combine iterative software development with a feasibility study in an introductory BFH mathematics course. For each exercise, Cairn estimates prior knowledge, assigns a support level using explicit rules, and asks the language model to generate only the corresponding hint. Automated checks prevent solution leakage, and each teaching decision is logged. We evaluate three questions: Does the diagnostic assign an appropriate support level? Are the hints well judged? Is the tool reliable and used in practice? Diagnostic results are compared with a short pre-test or, where available, Illuminate’s Bayesian knowledge-tracing (BKT) model. Lecturers assess sample hints against a shared rubric, while students report whether the support and fading pace match their needs.
Planned Translation
Transfer is built into Cairn from the start. The project plans to recruit two BFH mathematics lecturers to co-design exercises and use the web application in an introductory course. Their feedback and the pilot findings will guide development. Cairn will be released under a permissive open-source licence with a quick-start guide, documented API, open skill format and reusable example. Universities will be able to self-host it, connect their preferred AI model and integrate its tutoring capabilities into platforms such as Moodle. Results will be shared in a BeLEARN demonstration. Bridge & Binary will offer optional implementation, hosting and support, while BFH Virtual Academy plans to reuse the platform in Illuminate and future research.
Cairn aims to make personalised learning support practical for larger and more diverse student groups—without turning AI into an answer machine. Learners should receive the right help at the right time; lecturers and institutions should gain an inspectable, adaptable alternative to closed AI tutors. In year one, we will measure feasibility rather than claim learning gains. Indicators include successful course deployment and usage, agreement between Cairn’s support levels and an external diagnostic, lecturer ratings of hint quality, and student feedback on support fit and fading pace. Beyond the pilot, we will track partner integrations and reuse of the open-source software. The platform will also support later studies of learning effectiveness.
