AI Catalog Attribute Enrichment for Magento 2 by Maksym DernovyiAI Catalog Attribute Enrichment for Magento 2 by Maksym Dernovyi

AI Catalog Attribute Enrichment for Magento 2

Maksym Dernovyi

Maksym Dernovyi

Problem

A product shows up in layered navigation only when its filter attribute is filled. In many catalogs the value is written in the description while the attribute itself is empty, so shoppers who filter never see those products. Filling thousands of attributes by hand is not realistic.

What I built

A free, open-source Magento 2.4.9 module (PHP 8.3–8.5) that reads each product's name and descriptions and picks a value for empty select, multiselect and Yes/No attributes. It only chooses from the attribute's existing options and never invents text or new options. Answers come from the TypeSafe API, paid by the store directly.

Technical highlights

Each select and Yes/No question is asked twice with the options in opposite order; a value is accepted only when both answers agree above the threshold. "Not stated" is always an option, so silence is never stored as "No"
Rule modes: off, shadow (record only), review (admin approves or rejects) and auto. Every applied value can be reverted
A calibrate command measures precision on products that already have a value and suggests a threshold
Fills empty values only; store-view values, configurable variant attributes and values a human rejected are left alone
Cost control: unchanged products are not re-sent, identical requests are cached, monthly token budget
Runs from CLI (with --dry-run), the product grid or a nightly cron, through Magento message-queue consumers, with separate ACL resources for rules, suggestions and runs
No customer or order data is sent; only product text, attribute and option labels and rule text
README test: on one furniture store, an extra rule answered 96% of 149 products with a known "Number of seats", with no wrong values.

Link

Like this project

Posted Nov 1, 2026

Open-source Magento 2.4.9 module that fills empty filter attributes from product descriptions, choosing only existing options, with review and revert.