How do you train a chatbot on company documents?

Usually the model isn't "trained" from scratch: you build a knowledge base with approved documents, index the content, and use a retrieval system to pull the relevant passages before generating a response. You define instructions, citations, limits, tests, and a process for updating the sources.