Built independently by Szemat.Pro: this page is a curated index into the public documentation of the AI agent for regulated industries —its architecture decision records, its governance and its diagrams—, open and in three languages. It points to the sources rather than re-hosting them.
I am Waldemar Szemat, AI Solutions Architect and Fractional Head of AI. The full technical documentation is published and open; this page maps what it contains and why it matters, so it can be read with a guide.
What this page is
An index, not a copy. The live system has its own piece; this collects the what and the why of its design decisions and points to where each one is documented in full. Everything linked here is already public.
Architecture decision records (ADRs)
The ADRs record every structural decision of the agent —the separation of retrieval, reasoning and generation; guardrails on both input and output; observability treated as a requirement— with its context, its alternatives and its consequences. They include the records that ratified real corrections caught before production. The what and the why are public; the fine-grained how stays private. Read the ADRs.
Governance and compliance
The governance maps recognized frameworks —NIST AI RMF, ISO/IEC 42001, SOC 2, the EU AI Act, MITRE ATLAS, GDPR and Chile’s laws 19.628 and 21.719— onto the system’s decisions. It states design alignment, not third-party certifications. Healthcare stands here as one of the highest-bar industries, not as the identity of the work. Read the governance.
System diagrams
The diagrams show the agent’s flow and the evaluation gate that governs every promotion to production: no green verdict, no deploy. They are public copies, with no endpoints, API schemas or internal identifiers. See the diagrams.
What is not published
The probe corpus, the rubrics and the scoring logic of the evaluation harness are not published: they are the asset that does the work. What is tested and why is documented; the how never is. Any data shown in the material is 100% synthetic.
Accessing the live system
The documentation is open. The live system is coordinated through an access key, and every request is reviewed by a human; it is handled from the system piece, not here.
Publishing the what and the why —and holding back the how— is what makes an AI agent auditable without giving away the asset that makes it work.