AI Architectures · Research Infrastructure · Privacy by Design · Qualitative Data

I build digital infrastructure for research with sensitive data: AI-assisted analysis that runs on local models, anonymisation that fails closed rather than open, and metadata models that keep interview data usable across institutions. Local-first, GDPR-compliant, reproducible.

University of Luxembourg · ORCID · Website · Zenodo

DINOH — a digital, AI-assisted infrastructure for oral history. Multilingual benchmark corpus and evaluation suite: synthetic gold standard across seven languages, machine-readable schemas, WebVTT and OHMS export. Pipelines can be measured and compared before any real material is touched. v1.0.0 · MIT · 10.5281/zenodo.21273366

AegisQDA — local-first, fail-closed privacy gateway in front of DigQDA. Presidio-backed PII recognition, mandatory signed human review, typed document-local surrogates, second verification scan; data is released downstream only after sign-off. German, English, French; Luxembourgish optional. MVP · MIT

DigQDA — evidence-bound qualitative data analysis with local language models. Versioned method contracts, source-near coding prompts with machine-readable JSON schemas, deterministic source-unit segmentation, quote and locator validation, fail-closed conformance suite. pre-release (v0.4) · MIT

IMM-Core — minimal metadata model for interview-based research. Two-layer architecture (13 core fields, 7 mandatory, plus extensible implementation profiles), DCTAP-based, with crosswalks to Dublin Core, schema.org/Dataset, REFI-QDA/QDPX and CMDI. v1.0 · CC BY 4.0 · 10.5281/zenodo.20507328

Relational Justice Analysis — rule-based, fully traceable annotation framework for qualitative text data, with indicators for German, English and French. Every annotation can be traced back to the rule that produced it. v2.0.0 · MIT · 10.5281/zenodo.18640259

Released versions are openly licensed and archived on Zenodo with persistent identifiers. The DOIs above are concept DOIs — they always resolve to the latest release.

Local-first — processing happens inside the institution's own infrastructure. Sensitive data does not leave it, by architecture rather than by policy.

Fail-closed — where a system is unsure, it stops and asks. Silent guessing is the failure mode that makes automation unusable in this field.

Traceable — every result resolves to a location in the source and to the rule or method contract that produced it.

Interoperable — open standards and documented crosswalks instead of proprietary formats. The data has to outlive the tool.

Rather than automating interpretation, I design analytical architectures that preserve theoretical clarity, methodological transparency, and structural accountability. My broader aim is to operationalize sociological reasoning without reducing it to black-box computation.

The work rests on a socio-cybernetic understanding of society as a recursive system of distinctions — reproduced through feedback, institutional routinisation and cognitive stabilisation. From this follows a methodological commitment: analytical categories are not inferred from the data but theoretically grounded and made explicit.

Background in social and cultural anthropology; empirical work on social differentiation, migration, citizenship and oral history.