Nomula

An intake console that turns incoming requests into confirmed, routed work items. Each intake type is defined by a versioned profile that sets which fields to extract, the allowed categories and priorities, and the destination queues. A deterministic rules engine reads the request, suggests a category, priority and queue, and quotes the exact text behind each suggestion. A person then confirms or overrides it before anything is routed, and every decision is logged. Two intake streams (maintenance requests and open-ended feedback) run from configuration alone, with no code changes. An evaluation harness compares rules-only, model-only and hybrid methods on held-out synthetic data with pass thresholds set in advance. It is built for a local Llama model that only proposes labels from the configured list, must quote the request verbatim, and steps in only when the rules are unsure. Status: the rules baseline, review workflow and evaluation harness are complete and tested. The Llama integration has not yet been run end to end.