Hello-Wiki

Extract structured knowledge (summary, compiled truth, tags, temporal range) from text chunks during ingest Step 2. Uses domain-specific references that customize the output schema and prompt. The prompt is composed from the reference's system prompt plus the tag hierarchy injected into the user message.

Generate an initial hierarchical tag taxonomy for a knowledge base domain by calling LLM once. Uses domain-specific references (general, university-policy, etc.) that customize the generation prompt. The output maps to the tags table (ltree paths). The generated 2-level tree is the starting point — deeper nesting emerges naturally during ingest as the extraction LLM splits overcrowded leaf tags into sub-leaves.