How can learning platforms such as Moodle benefit from modern AI - without losing data protection, source validity or pedagogical control? This question is at the heart of a research grant-funded project by eLeDia GmbH, which combines the Learning Management System Moodle with a secure, locally operated AI architecture.
The development of a Retrieval Augmented Generation (RAG)-based infrastructure and the eLeDia.ai Tutor creates a system that not only generates didactic content automatically, but also makes it accessible to learners in an interactive, context-related form - without a cloud, without data outflow, but with maximum didactic precision.
Why research is so important for SMEs
Research in the field of artificial intelligence is often associated with large corporations. However, it is precisely small and medium-sized enterprises (SMEs) that produce particularly practical innovations thanks to their proximity to the market, agility and technical expertise.
With the research allowance under the Research Allowance Act (FZulG), Germany has created an instrument that promotes precisely this spirit of innovation. SMEs can use it to claim up to 35% of their eligible research and development expenditure against tax - regardless of industry, topic or project size.
For eLeDia, the research grant is a catalyst for developing forward-looking technologies for the educational landscape - open source, transparent and data sovereign.
Research objective: Didactically meaningful use of AI
The project is pursuing a long-term vision: the AI-based, valid and data-secure creation of teaching and learning materials for universities, further education providers and companies.
Currently, course materials are usually created manually - a process that can take weeks. Although existing AI tools provide draft texts, they hardly meet the requirements of professional educational environments:
- no guarantee for correctness of content,
- no protection of sensitive data,
- No integration into systems such as Moodle.
This is what the project is all about: using a capsuled, locally operated AI system based on Ollama and Moodle, a system is created that supports teachers in the creation of course content, exam questions or training documents - quickly, securely and comprehensibly.
Retrieval augmented generation as core technology
At the heart of the project is RAG technology (Retrieval-Augmented Generation) - a method that combines generative AI with structured knowledge retrieval. While traditional LLMs access their own, often opaque model knowledge, RAG works with two clearly separate steps:
- Retrieval: Relevant course information is extracted from Moodle, semantically analyzed and stored in a specially developed n-dimensional vector database.
Access rights and permissions are taken into account - the system knows exactly what content a person is allowed to access. - Augmented Generation (generation): The LLM then accesses this verified knowledge base to generate new course materials, tasks or explanations - always with reference to the original source.
The result is an AI system that works in a context-aware, data-secure and verifiable manner - a basic requirement for trustworthy AI in teaching.
The eLeDia.AI Tutor: A tutor directly in the Moodle course
The eLeDia.ai Tutor builds on this architecture - an interactive learning assistant that makes the content of a course accessible to learners in an alternative, dialogic form.
The tutor makes the learning content of a course available via the chat. The tutor explains content using the original sources from Moodle. It transparently refers to references in the course on which the explanation is based.
The result is a didactically valid, context-aware learning assistant that does not replace knowledge, but makes it understandable. It expands the concept of digital learning with an intelligent, secure level of interaction that can be used by universities, further education institutions and companies alike
Technological and scientific innovations
The project comprises several independent research and development strands:
- Data-secure RAG architecture: Completely open source (Ollama, Moodle, vector database), can be operated locally, without external cloud services.
- Adaptive vector database: Dynamic modeling of Moodle permission structures for granular access control.
- Container-based orchestration: Kubernetes-based environment for load scaling and performance optimization.
- Development of a Moodle plugin: Access to the eLeDia.ai tutor directly in Moodle.
Research as a strategic foundation
One challenge for the project lies in the balance between technical openness and data-secure control. The system developed must be flexible enough to understand different Moodle installations on the one hand, and precise enough to guarantee data sovereignty and validity on the other. However, it is precisely this technical uncertainty that makes the project worthy of funding - and highlights the importance of risky research in SMEs. The research grant creates the space to break new ground without the economic risk of blocking the entire development.
The prototype has already been published and is being tested.
The combination of RAG technology, local AI integration and the eLeDia.AI Tutor creates a product that further develops the way in which learning content is created and taught. The project shows that research is not a privilege of large corporations - it is also worthwhile for SMEs like eLeDia: the research grant makes research visible, feasible and sustainably effective.
AI transparency note: For this blog post, the multi-page research proposal was converted into shortened and target group-specific texts with the help of AI and then editorially revised.
