Bringing Artificial Intelligence (AI) into educational practice prompts critical reflections on teaching methods, assessments strategies, and learners agency. This paper centers on the design, implementation, and deployment of DHAgent, a domain-specific Large Language Model (LLM) embedded within the undergraduate Digital Humanities (DH) curriculum at the Department of Languages and Cultures, Western University (London, Canada). Amid growing concerns about the uncritical adoption of generative AI in educational contexts, this initiative rethinks the DH class as a place for hands-on experimentation between students, instructors, and machines, positioning the AI agent as both a conversational counterpoint and a creative partner in the interpretive process.

The teaching project aims to address two main questions:
(1) how to align AI behavior with the interdisciplinary, non-deterministic ethos of the DH, and
(2) how to encourage student agency and literacy in the critique, and ethical use of AI systems?

The methodological foundation of the DHAgent relies on Retrieval-Augmented Generation (RAG), a technique that combines text generation with dynamic access to a curated external knowledge base. In this case, we leveraged a training corpus of over 50,000 anonymized forum posts, generated between 2019 and 2025 by students enrolled in six DH courses. The dataset encapsulates discipline-specific discourse, vocabularies, and interactions often absent from mainstream LLM training data. In class, the model was not introduced as a finished tool but as a co-designed cognitive artifact. Students participated in iterative feedback rounds involving guided LLM interactions, AI output critique, and reflective pieces. These activities are embedded into classroom practices through structured weekly learning cycles, and we placed emphasis on transparency, bias detection, and critical engagement.

The paper describes two principal outcomes of this project. First, it provides a detailed design blueprint for developing discipline-specific LLMs grounded in contextualized, niche knowledge. Second, it introduces a participatory pedagogical model that structures AI use around critical dialogue, student co-authorship, and iterative refinement. We also show classroom scenarios and use-cases that demonstrate how LLMs can be incorporated not as mere content providers, but as stimulus for interpretive thought, challenge assumptions, and deepen engagement with course material, without displacing human judgment or creativity.

We argue that engaging students as co-designers and reflective users of AI, can bridge digital competence with humanities values and foster a generation of learners who not only use AI but understand and shape it.

Keywords: Artificial Intelligence, Large Language Models, Learning Methodologies, Students Agency, AI Literacy.