Tracking Regulation with an LLM Wiki

Overview

Part of my job involves tracking gaming regulation across more than a dozen markets: new laws, regulator guidance, enforcement actions, and how other publishers respond in public. I built a legal-regulatory wiki to organise that material into something I can query: not just a record of what happened, but a retrieval layer for precedents, comparisons and regulatory patterns.

The wiki is inspired by Andrej Karpathy’s LLM Wiki pattern. Instead of asking an AI tool to search online afresh, or reprocess the same raw materials every time a question comes up, the model helps maintain a persistent, interlinked set of Markdown pages. New sources go in; Claude extracts the relevant obligations, regulatory developments and industry responses, then files them under the right jurisdiction and topic.

The source base is public by design: legislative text, supplementary guidance published on regulator websites, regulator announcements, enforcement notices, news articles reporting on regulator press conferences or enforcement developments, public announcements from gaming companies, and news articles reporting on measures adopted by companies in response to regulation.

The wiki sits in an Obsidian vault, which in my setup is a local folder of Markdown files stored on my device and not uploaded to the cloud. Obsidian provides the reading, linking and navigation layer. Claude reads and writes the underlying .md files. An instruction file, such as CLAUDE.md, controls how sources are ingested, how pages are updated, how cross-references are maintained, and how query answers cite back to the underlying public materials. The result is a local, public-source knowledge base that I can browse manually in Obsidian, or query with an LLM when I need a structured answer.

My Use Case

The value shows up when the question is practical rather than abstract. In gaming regulation, the question is often not just “what does the law say?”, but “what are regulators saying, how are companies responding, and what options appear to be emerging?”

Brazil’s Digital ECA is a good example: it bans paid loot boxes in any game minors can access (effective 17 March 2026). I had already ingested sources on how gaming companies responded, from their own announcements to news coverage, so I could ask the wiki directly what actions they took, and get the answer back with concrete examples, each linked to its source.

The wiki also helps with early pattern recognition. When several markets move towards age assurance, child-safety impact assessments, loot box restrictions or risk-based minor-protection frameworks, the developments no longer sit in separate files or chat histories. They start to form a linked regulatory picture.

Back to AI Journal