His goal wasn’t a flashy chatbot. It was infrastructure: a system that knows its sources, ranks them by legal quality, and won’t confuse commentary with binding law.

2025-11-28 · Source: Pascal’s Substack · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Compliance & Risk Management · Depth: Advanced, medium

Summary

Oliver Belitz developed a "second brain" for the EU AI Act, a structured knowledge base costing approximately €60 and 2.5 million tokens, designed to provide reliable AI assistance in compliance-heavy domains. Unlike traditional RAG systems that retrieve isolated snippets, this infrastructure knows its sources, ranks them by legal quality, and identifies conflicts without fabricating certainty. Belitz's system, inspired by Andrej Karpathy and built with Claude Code and Obsidian, uses a maintained markdown corpus with explicit cross-links, avoiding vector databases. Key features include source hierarchy, collision logic, statutory grounding, anti-anchoring rules, and semantic drift control, ensuring the AI behaves like a disciplined research function. Oliver Schmidt-Prietz validated this approach, emphasizing that token burn is scaffolding, wikis require policing against drift, and separate "regulatory" and "general" vaults are crucial to prevent contamination.

Key takeaway

For CTOs or VPs of Engineering building AI systems in compliance-heavy or high-stakes domains, you should prioritize pre-compiled, governed knowledge infrastructure over ad-hoc retrieval. Your teams must budget for upfront token costs as scaffolding, implement rigorous maintenance to prevent semantic drift, and establish separate knowledge vaults for different levels of truth. This approach enhances factuality, auditability, and conflict handling, reducing reputational and compliance risks by providing transparent, authority-respecting AI outputs.

Key insights

Curated, structured knowledge bases with explicit governance outperform naive RAG for high-stakes AI applications.

Principles

• Uncertainty belongs in the answer.
• Structure knowledge by authority hierarchy.
• Audit knowledge graphs periodically for drift.

Method

Build a knowledge graph using a maintained markdown corpus with explicit cross-links, separating sources by legal quality, implementing collision logic, and requiring statutory grounding for answers.

In practice

• Separate regulatory from general knowledge vaults.
• Implement anti-anchoring rules for prior publications.
• Design periodic audits beyond basic lint checks.

Topics

• EU AI Act
• Knowledge Graph Architecture
• Legal AI Systems
• Source Hierarchy
• Semantic Drift Control

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, AI Architect, Legal Professional

Related on AIssential

• The Invisible Handshake: How We Are Accidentally Teaching AI Systems to Agree with Each Other — AI systems inadvertently converge on shared perspectives when users chain their outputs, creating an "invisible handshake" of consensus.
• Guiding a Safe Future for AI – Part 2 — AI's rapid advancement necessitates a focus on safety, ethical deployment, and robust infrastructure to serve humanity.
• Controlling Authority Retrieval: A Missing Retrieval Objective for Authority-Governed Knowledge — Controlling Authority Retrieval (CAR) addresses knowledge validity in authority-governed domains, outperforming semantic search.
• Semantic drift and semantic integrity: Stewarding meaning in the age of AI — AI exponentially accelerates semantic drift, demanding a shift to continuous semantic integrity to govern system meaning.
• Communication Isn’t the Problem. Retrieval Is. — Organizational failures stem from retrieval breakdowns, not knowledge gaps, where information exists but cannot be accessed effectively.
• When accurate AI is still dangerously incomplete — In high-stakes domains, AI requires "completeness" and authority beyond mere accuracy and relevancy.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Pascal’s Substack.