A developer has shared a pattern for building personal knowledge bases using LLMs that moves beyond traditional RAG systems. The approach involves creating a persistent wiki where the LLM incrementally builds and maintains structured markdown files rather than rediscovering knowledge from scratch with each query. This wiki serves as a compounding artifact that gets richer with every source added.

Unlike conventional systems like NotebookLM or ChatGPT file uploads that retrieve from raw documents at query time, this method has the LLM extract key information and integrate it into existing wiki pages. The system maintains cross-references, flags contradictions, and keeps the synthesis current without re-deriving everything on each question. The architecture consists of three layers: raw immutable sources, the LLM-generated wiki, and a schema document that tells the LLM how to maintain the knowledge base.

Users interact with the system by sourcing documents and asking questions while the LLM handles all summarizing, cross-referencing, and maintenance work. The approach works for various contexts including personal tracking, research, business wikis, and competitive analysis. Tools like Obsidian serve as the interface while the LLM acts as the programmer maintaining the wiki codebase. This represents a shift from retrieval-based systems to persistent knowledge accumulation.