Product Database with Automated Claim Checking

Status: in use · PostgreSQL · Python · WooCommerce API · Telegram bot

The problem #

Product information was spread across the shop, a grown series of Excel versions, CSV exports and Notion. At the same time, health-related claims on food are strictly regulated in the EU — one wrong sentence on a product page is a warning-letter risk.

What I built #

The database mirrors the shop into a single place, with versioned copy and ready-made views for the team.

The actual core is the checking layer. The health-claims rules exist as a table, a scanner tests the live pages against them, and a bot that lets shop copy be written by chat is only allowed to publish after that check has passed.

Why it is built this way #

The model writes; the rule decides. A language model is good at phrasing a product description and bad at knowing which phrasing is legally permitted. Separating those two jobs means the compliance decision stays in something a human can read, audit and correct — a table of rules, not a prompt.