A Governance-First Architecture for AI-Assisted Qualitative Research

The proliferation of large language models as analytical tools in qualitative research has arrived faster than the governance frameworks needed to deploy them responsibly. Existing platforms prioritise capability while leaving the questions that define whether an analysis is credible — reflexivity, auditability, data protection, evidential grounding — for the researcher to solve manually. This paper describes a governance-first architecture for AI-assisted qualitative research: a modular, open-source pipeline in which methodological safeguards are enforced at the infrastructure level rather than delegated to researcher self-discipline. The contribution is architectural rather than algorithmic: the pipeline specifies a governance infrastructure within which AI-assisted analysis can be conducted responsibly, regardless of which underlying model is used. The architecture achieves this through a two-pass analysis structure with cryptographic anchoring to prevent post-hoc revision, a structured researcher reflexivity elicitation module locked before the second pass, DPIA gating as a hard pipeline constraint for special-category data, per-claim grounding verification with human sign-off blocking synthesis, and an automatically generated authorship disclosure record. Argued here is that enforcing these safeguards programmatically — rather than recommending them procedurally — constitutes a meaningful advance in what responsible AI-assisted qualitative research can look like. The pipeline is presented as an untested design contribution; the repository is publicly available for inspection and empirical evaluation.