LLM wiki à la Karpathy – context engineering in practice
Context engineering as a concept: a personal, growing wiki as a local context source for LLM queries – structured instead of naive vector search.
Notes, learnings and research were scattered across loose files. To feed an LLM meaningfully with my own knowledge, what was missing was a structured, continuously growing source instead of isolated snippets. Concept and implementation – data model, backend and local LLM integration – were mine, single-handedly.
Tech stack: TypeScript · PostgreSQL · Qwen3-32B · Ollama
A wiki following the Karpathy pattern
Knowledge is accumulated continuously and stored in a structured way in PostgreSQL, TypeScript is the end-to-end stack, and Qwen3-32B via Ollama handles the local LLM features. The focus is on context engineering – the wiki feeds precisely the right context into the model instead of relying on naive vector search.
The result
A personal, growing knowledge base that serves as a local context source for LLM queries – fully offline, no cloud.
What I took away from it
Context engineering works in practice: structured, accumulated notes beat naive vector search. The value comes from consistent upkeep – a wiki is only as good as the discipline behind it.