Engineering engagements

    Article 50 and Articles 12–14, built into your system

    Legal advice tells you what the EU AI Act requires. It doesn't build the disclosure, marking, logging and oversight into your code. That's the part I do, while the Aevoxis product is in development. We start with Article 50, which already applies, and extend the same components to Articles 12–14 before they apply in December 2027.

    Article 50 transparency and content marking

    Article 50 has applied since 2 August 2026, so this is the work most teams need first. Each duty becomes a change in the product, not a paragraph in a policy document.

    • Art. 50(1): AI-interaction disclosure built into the product, with a log that it was shown
    • Art. 50(2): machine-readable marking of AI-generated content at generation time
    • Art. 50(4): deepfake disclosure in the publishing flow

    You get the disclosures in the product, the marking applied at generation time, and a record that each one was shown.

    Gap analysis for Article 50 and Articles 12–14

    I go through your AI system's architecture and check it against the obligations that bind it. Is it clear to people that they are talking to AI, and is generated content marked? Where does logging stop? What can a deployer actually see? Can a person stop the system, or only watch it?

    You get a written gap analysis mapped to specific articles, ranked by risk and effort.

    Automatic logging (Article 12)

    I design and build event logging into the execution path: the input that triggered a decision, the decision, the model version and enough context to reconstruct it later. Records are hashed, so tampering is detectable.

    You get an audit trail an assessor can follow, produced by the system itself.

    Transparency for deployers (Article 13)

    I surface how an output was produced in the interface your users already work in, so a non-technical person can judge whether to rely on it.

    You get explanations where decisions happen, plus the instructions for use the article requires.

    Human oversight (Article 14)

    I build oversight into the critical path: thresholds that stop the system, a decision point a person can't skip, and override that actually overrides.

    You get working oversight that holds up when someone asks you to demonstrate it.

    Who I work with

    Small and mid-sized companies and scale-ups in Germany and the EU that build or deploy high-risk AI, in any industry. Engagements are remote. I'm one engineer, so I take on a small number at a time.

    • Google Developer Expert

      Machine Learning

    • AWS Community Builder

      Machine Learning

    • AWS AIdeas 2025

      Top 300 of 10,000+ entries

    • M.Sc. Data Science

      Newcastle University

    • IEEE and Elsevier

      Invited peer reviewer

    • Published author

      Apress

    References available on request.

    Questions

    What does an engagement start with?
    A gap analysis. I go through your system's architecture and map what it does today against Articles 12, 13 and 14. You get a written list of gaps, ranked by risk and effort.
    Do you work on existing systems or only new ones?
    Both. Most teams already have a system in production. Logging and oversight can be added to it, but some decisions are cheaper to change early, so the gap analysis also says what to fix first.
    Which tech stacks do you work with?
    Python-based AI systems, including LangChain, LangGraph and CrewAI pipelines, on AWS (Lambda, Bedrock) or self-hosted open-source models. Other stacks are possible; ask.
    How are engagements invoiced?
    Through SMartDE eG, a cooperative for self-employed professionals in Germany. Its details are in the Impressum.

    Not sure where your system stands?

    Answer this five-question technical check. I'll review your answers personally within 24 hours.

    Tell me about your system

    Describe what it does and when you need to be ready. I'll tell you honestly what to tackle first.

    Get in touch