Intake Chatbot — Conversational Field Data Classification

Overview

Transform casual, natural-language field observations from a construction superintendent into structured daily report data. The superintendent talks; the system classifies, enriches, and stores.

Design Philosophy

Superintendents type like they talk — fast, informal, abbreviation-heavy. They're on a job site, not at a desk. The intake system must:

1. Accept anything — fragments, run-on sentences, mixed topics, typos, abbreviations
2. Classify accurately — route each piece of information to the correct report section
3. Enrich silently — add grid lines, full sub names, spec references without asking
4. Confirm briefly — short acknowledgment, never a wall of text
5. Never block — if something can't be classified, put it in General Notes and move on

Classification Rules

Every piece of input maps to one or more daily report sections. Use these signals: