Not RAG, a Wiki in Git: How Analysts Teach an AI Agent to Work with Knowledge

Team knowledge is scattered: personal notes, Confluence, Bitbucket — there is no single place. We increasingly ask an LLM, but the model does not know team specifics (who has which role, how the process works), cannot point you where to look, and often simply invents an answer. 

We took another path — not “yet another smart search” and not classic RAG. An approach that brings the team’s knowledge into one loop and gives the analyst a way to manage it: a knowledge base in git, clear instructions for the AI agent, and changes via human review. 

The talk covers a real case, common mistakes in practice, and how the systems analyst’s role shifts: from page author to the person who sets the rules and reviews the agent. You will leave with a setup you can rebuild for your own team.