This work presents the design and implementation of an educational chatbot powered by generative AI. The system employs a retrieval-augmented generation (RAG) architecture that integrates large language models (LLMs) to deliver accurate, contextually relevant, and pedagogically sound responses. Unlike general-purpose chatbots, this solution prioritises three core features: pedagogical behaviour informed by learning theories; the exclusive use of validated open-access academic sources; and the inclusion of verifiable references in each interaction. This approach reduces the hallucinations that are common in generative models and enhances the didactic reliability of the responses provided. Pedagogical principles such as constructive alignment and scaffolding inform the chatbot’s interaction model, enabling it to provide adaptive explanations and engage in Socratic dialogue. The project has resulted in a validated prototype in the specific domain of Generative Artificial Intelligence in Education, which demonstrates three main capabilities: improved conceptual understanding through interactive dialogue, reduced factual errors compared to standard chatbots, and a customizable knowledge base supported by a technical implementation guide. The system was developed in Python using LangChain and Milvus, and is released as open source to promote transparency, adaptability, and replication in educational contexts. Finally, this work provides a replicable model for integrating LLMs into educational chatbots informed by pedagogy.