This proposition reports on a series of pedagogical interventions conducted with approximately thirty students enrolled in a Master’s program in Humanities and Cultural Industries. Initially structured around two technical case studies — Retrieval-Augmented Generation (RAG) as a writing companion and AI-based image annotation for visual humanities — the course addressed contemporary transformations of textuality, authorship, and editorial mediation in digital environments. These technical topics, however, rapidly became entry points for a broader reflection on research practices, pedagogical formats, and interdisciplinary knowledge production.

The pedagogical framework combined inquiry-based learning and critical AI literacy. Rather than maintaining a vertical model of expertise transmission, the sessions were shaped by continuous exchanges between students’ humanities training, their personal experimentation with generative systems, and the instructor’s background in computer science. This dialogical configuration progressively transformed the classroom into a shared methodological laboratory.

The research contribution consists in an empirically grounded analysis of how interdisciplinary teaching can support methodological innovation in digital humanities infrastructures. The course was organized as iterative cycles of experimentation, collective discussion, and methodological redesign. Two main conclusions emerged from these interactions.

First, students’ proposals — sometimes perceived as naive from a strictly technical perspective — proved to be methodologically fertile precisely because they were not yet constrained by disciplinary automatisms. In RAG-based writing experiments conducted in journalistic and mediation contexts, students used contrasting document corpora to explore how source selection and qualification shape narrative production. These experiments challenged the assumption that methodological innovation primarily originates from senior research communities, and highlighted student contributions as legitimate research inputs.

Second, the collective discussions fostered a reflexive posture regarding research objectives, methodological choices, and epistemic limits. In the case of RAG, students identified how corpus composition, document status, and prompt framing condition narrative outcomes, positioning RAG as an epistemic device rather than a mere generative accelerator. In the case of annotation, they critically examined how current pipelines impose reductive representations of visual objects, echoing critiques of classification as an epistemic construction. Students proposed differentiated annotation strategies depending on research objectives, combining automation with humanities-driven interpretative control.

RAG and annotation therefore function not as the core objects of this contribution, but as methodological catalysts for rethinking how digital humanities knowledge is produced, evaluated, and transmitted.

The paper proposes a reflexive circular framework: (1) appropriation of emerging methods; (2) critical examination of their assumptions and associated methodological conventions; (3) collective proposal of alternative methodological configurations; and (4) reintegration of these proposals into both research and teaching practices. Methodology is understood broadly, encompassing research questions, evaluation criteria, pedagogical design, and epistemological and ethical positioning.

From a DARIAH perspective, this experience advocates for more participatory, horizontal, and research-integrated pedagogies in digital humanities. It argues for stronger connections between teaching and research, and for the inclusion of Master’s students as active contributors within participatory research infrastructures. Such configurations challenge the traditional division between engineers as technical executors and humanities scholars as users, promoting instead a genuinely interdisciplinary co-construction of future digital humanities methods.