EuConform

I built EuConform out of frustration with how abstract AI compliance often feels in practice. With the EU AI Act approaching, I kept seeing the same gap: lots of legal interpretation and high-level checklists, but very little guidance on how to translate those requirements into concrete, inspectable technical steps. Most existing approaches also assume cloud-based analysis, which felt problematic for sensitive models, datasets, or internal systems. My goal with EuConform was to explore a local-first, transparent approach to compliance: - "turning risk classification into an explicit technical process" - "making bias evaluation measurable (not just declarative)" - "generating documentation directly from the analysis, not as a manual afterthought" During development, the focus shifted from “cover everything” to making assumptions explicit: showing how a result was derived, when exact probabilities are available, and when fallbacks are used. EuConform is a personal open-source project and very much a work in progress. I’m mainly hoping to learn whether this approach makes sense in real-world projects, what feels missing, and what feels unnecessary. Happy to hear any feedback, criticism, or ideas.