Ai · Technology

Two Failed Builds, a Conversation, and the Knowledgebase That Came Out of It

A client had already tried twice to build an AI-driven knowledgebase before I got involved: a vibe-coded custom RAG system first, then a Custom GPT setup. Both fell short in different ways. The fix wasn't a third rebuild, it was a conversation about what had actually gone wrong, and the taxonomy-first architecture that came out of it, now called Smart Knowledgebase.

Login screen showing three users with different roles and access levels Search interface with retrieve and answer modes, and a full-corpus or auto-scope toggle Document ingestion form with sensitivity, department, and explicit tag and category fields Pending tags review panel for approving, rejecting, or merging LLM-proposed tags Taxonomy management screen showing categories and tags as a managed vocabulary Knowledgebase selector grouping scopes into named collections

The full write-up, including what went wrong the first two times and what actually fixed it, lives as a case study found at

Weft & Whetstone Solutions & Strategies: Smart Knowledgebase

← All thoughts