Knowledge base automation with RAG
A wiki or share drive becomes answerable only after conversion is automated; chat UI without ingest automation is theater.
The knowledge base already exists
Confluence exports, Notion dumps, shared drives, ticket archives - teams ask for 'a chatbot over our KB' as if the KB were waiting for a model. It is waiting for conversion. Automating the KB for RAG means every published page or file lands as passages on a cadence you chose.
UI without ingest is a costume
A polished chat box over a stale or empty index trains users to distrust RAG. Automate convert first; ship the chat when the corpus is measurable. Citation metadata - source and page - belongs in the chunks so answers can point back at the KB entry, not at vibes.
Structure that survives export
Wikis become HTML or Markdown; drives become Office and PDF. Each format has different yield. Automate per source type with honest skips for empty scans. A single 'ingest everything' hammer that indexes boilerplate navigation text fills the store with noise.
Ownership and access
KB automation inherits ACL problems. Prefer converting paths the answering app is allowed to show. Local convert for restricted spaces; API convert for published spaces. Do not flatten confidential and public trees into one collection and hope retrieval respects politics.
Measure freshness
Track last successful convert per source, files processed, files skipped. A KB that looks automated but whose last run was three weeks ago is manual RAG with better branding. Dashboards beat anecdotes.