Every grounded AI feature is only as good as its knowledge base: if the price list is stale, the model quotes stale prices fluently. The work is unglamorous — deciding which documents are authoritative, keeping them current, chunking them so the right fragment can be found — and it is most of the project.
A good one carries metadata as well as text: which product, which language, valid from when. Those fields are what let a system answer 'what does it cost' with this year's price instead of a confident average of every price ever written down.
Related terms
RAG (retrieval-augmented generation)
Looking up relevant documents first and putting them in the prompt, so the model answers from your material instead of from memory.
AI chatbot
A conversational interface backed by a language model — useful when it answers from your real data and hands off gracefully when it cannot.
Vector database
A database built to find the stored items whose embeddings are closest to a query's, quickly, across millions of rows.
The bench this belongs to
AI automationThe repetitive half of your week, handed to software that does not get bored. Inbox triage, follow-ups, reporting, data entry between tools that were never meant to talk.
