Pri's wroclove.rb 2026 lightning talk. Presents Arkency's new open-source 'internal Wikipedia' tool: a Ruby application that ingests transcripts and uses an LLM to extract a knowledge graph defined by a pluggable ontology (note kinds + typical relations between them). Demoed by feeding the last six editions of wroclove.rb transcripts, producing a public graph reachable via QR code. Each ingestion is a transcript; extraction attempts are tracked with token usage, time, LLM call counts and round-trip counts. Example extraction of Szymon Fiedler's 2025 talk produced 28 new nodes and 49 new edges. Bonuses called out: full audit log + event-sourced graph, and a built-in MCP server so any LLM can query the graph (asked 'what was the most trending subject last year?' the MCP returned 'SQLite'). Entity resolution is achieved by exposing three Ruby-LLM-style tools (list_nodes_by_kind, search_nodes, get_node_edges) to the LLM during extraction so repeated mentions of the same entity (e.g. Andrzej) across multiple conferences resolve to a single node. The only thing needed to reuse the codebase on another domain is to define an ontology.