Fix your architecture before it fixes you.

I build the data architecture that EU AI Act compliance actually requires. Public reference implementations. Working YAML contracts. SQL that survives the audit. Currently focused on the multi-entity joint deployer pattern — the structural problem nobody productized yet, the one every group structure hits the moment a subsidiary runs the platform the parent procured.

Article 10 implementations. The EU AI Act requires documented data governance for high-risk AI systems. Most published guidance assumes a single deployer. Real organisations are not built that way. I build the contracts and the SQL that close the gap.

Joint-deployer YAML. When two legal entities share authority over the same AI system, both are deployers under Article 3(8). Both owe Article 26 obligations. The contract structure that handles this — composable rows per entity, joint responsibilities, explicit hybrid scope — is the work. Reference implementation public on GitHub.

Evidentiary data contracts. Most data documentation is policy text in Confluence. None of it survives a regulator inquiry. I write contracts that double as legal evidence — the same artifact answers "is this row correct?" and "could you prove that to a supervisor?"

The reference implementation. A governed star schema data warehouse built from 39.5M rows of EU energy data in 8 weekends. YAML data contracts written before a single SQL query.

Read the multi-entity pattern document — the dilution problem, the CJEU analogy, the four-block composable pattern, three twelve-month failure modes.

Architecture First — one architecture rule every Friday, 8:00 CET. Beehiiv-first, with shortened LinkedIn newsletter teasers. Topics:

• The multi-entity joint deployer pattern — when single-deployer YAML breaks
• Append-only by database role (the audit log pattern that survives regulator inquiry)
• 88% over-count: how naive row counting destroys Article 10 compliance estimates
• The renewable share error — caught, corrected publicly, content recovered

Subscribe to the newsletter

DuckDB. PostgreSQL. Power BI. Git. SQL. YAML (ODCS v3.1.0). PowerShell. Hand-drawn architecture sketches on actual paper.

LinkedIn DM: linkedin.com/in/michal-mrugala02

I do not publish a calendar link or email here. The newsletter and the GitHub repository are the right entry points.

_{Architecture First — peer-level reference implementations for EU AI Act Article 10 data governance.}