Tag
A practitioner shares the architecture for trustworthy quoting agents: the LLM only extracts specs from messy customer messages at temperature 0, nulls become follow-up questions, pricing is done by a deterministic code engine with regression tests, and a human approves every quote before it goes out.