Designing digital learning courses with generative AI

Piazza Copernico, a company in the e-learning sector, approached GRID+ to collaborate with the R&D Department on the implementation of an instructional design system based on generative AI. The project focused on the implementation of software that exploits Large Language Models (LLM) to automate course design and optimise training processes. This made it possible to significantly reduce the production time of educational content, while maintaining a high quality and customisation of the training material.

Piazza Copernico, as part of its AI strategy, decided to introduce artificial intelligence into the process, making the activity of analysing and designing semi-finished projects to be finalised with the content experts in storyboards more effective. In the process of creating courses and learning materials, it is necessary to work with accuracy and confidentiality on the materials provided by the client, in a process that starts with content analysis and proceeds with the creation of a macro-design of learning objectives to proceed with the reworking of the content to make it effective for representation in digital learning objects.

The complexity of L&D (learning and development) work requires instructional designers to generate content quickly, while maintaining subject specification and high design standards. The challenge in the introduction of artificial intelligence was to include tools to support the design phase, not only aimed at better time management, but also at maintaining design standards and an adequate representation of all the content provided while at the same time retaining a level of supervision of the result produced.

GRID+ collaborated with Piazza Copernico to develop a customised solution based on Generative AI. The designed system uses LLM with Retrieval-Augmented Generation (RAG) methodology to generate complex and customised texts for each course.
The real strength of the solution lies in its modularity: currently the open-source Llama 3.1 model is used as LLM, but the system has been designed to allow the quick and easy integration of future LLMs, keeping the platform up-to-date with the latest technological innovations. A further advantage of the application lies in its flexibility: in addition to the generation of teaching texts, it can be easily adapted to a wide range of tasks, such as the creation of content for audio lectures or as a search engine within the company's knowledge base. The system, which includes both the RAG and LLM models, is hosted entirely on dedicated local servers, without relying on external APIs, guaranteeing efficiency, robustness and respect for privacy. Data reside on Piazza Copernico's proprietary servers, ensuring maximum confidentiality in the management and transmission of information. The conversational user interface is designed to be intuitive and accessible, allowing instructional designers to interact with the system without the need for advanced technical skills. With simple inputs, instructional designers can customise content and obtain ready-to-use learning materials quickly and automatically. The system is optimised to handle a variable number of users simultaneously, ensuring efficiency even on a large scale.

The solution, currently in use within Piazza Copernico, is being tested and the impact on the work process is being measured. The important perceived benefit is the possibility of support for trainers in the analysis and drafting phases. The possibility of designing through data interrogation is important because it allows access to information sources at any time to verify the processing in a virtuous process of human-computer interaction. This allows instructional designers to increasingly focus on teaching strategies, while the generation of materials is largely automated.