A rule-based filter applies explicit rules to accept, reject, or route inputs or outputs. In conversational systems, it acts as a guardrail around otherwise open-ended responses. section 12.4
Rules provide useful control but cannot capture everything people do in conversation. Production chatbots therefore commonly combine filters with large language models and retrieval-augmented generation to reduce misleading or harmful responses. section 12.4
Guarding a chat product
A chat product can route user inputs or generated outputs through explicit filters before or alongside retrieval-augmented generation and large language model responses. section 12.4
