Abstract:

In recent years, the emergence of large language models (LLMs) has marked a turning point in the development of educational technologies that use artificial intelligence. The ability of LLMs to generate digital content such as text, images, and videos, among others, has transformed the processes used to produce, analyze, and personalize educational materials, making it possible to create more dynamic learning experiences that are adaptable to each student. In this context, the automated generation of educational resources represents a strategic opportunity to enhance programming education, allowing for the design of more relevant and personalized materials. Among the most widely used LLM models are OpenAI's GPT, Google's Gemini, and Anthropic's Claude 3, each with outstanding features and commonly available even on our mobile devices.

In the field of education, the implementation of various technological resources has led to the diversification of teaching methodologies. This transformation has required teachers to adapt their pedagogical practices and resort to the integration of technology to enrich educational environments and facilitate the acquisition of skills and knowledge by students. In light of this, COVIA was created as a tool to assist teachers in the field of programming education.

This paper presents the process of creating two types of educational resources using Generative AI: programming exercises and educational videos. The exercises were designed using GPT-5 and Gemini-2.5-pro, while the videos were generated with HeyGen. Both resources were integrated into COVIA. The programming exercises were created using a prompt that included the context and information of the lesson and the level of difficulty associated with the exercise. This request was sent to an LLM, which generated the exercise in response. The exercise was then reviewed, adjusted, and approved by a teacher to be stored as a resource within the COVIA database.

To generate the educational videos, HeyGen was first provided with a script containing written information about the lesson, including definitions, syntax, code examples, etc. Next, an avatar was selected and certain visual aspects were configured, such as music and other design components. Once all the elements and configurations were complete, the video was produced. Each video was stored on a YouTube channel created specifically for COVIA.

A total of 33 exercises were generated, and of the 33 videos available on COVIA, 10 were generated with HeyGen. During this semester, COVIA supported more than 100 university students in developing programming skills. Version 2 of COVIA is currently in development. In this new version, the exercises presented to students will be adapted to each student's personality.

Keywords:

Generative AI, Video-based Learning, Programming, Educational Platform, Learning Resources.