Our summary
A shop's descriptions may suggest that an item is a gift or suits a certain audience without storing those details separately. This recipe asks Jev to identify categories, audiences, and tags, then saves its answers alongside the unchanged product records.
The program checks which products still need labels before sending requests. It stores categories, probabilities, the model used, and the indexing date in separate tables. The database can then combine those labels with exact conditions, such as a maximum price, without asking AI to do arithmetic.
Some labels in the author's sample were debatable or wrong. A saved probability helps someone find uncertain records but does not prove the other labels are correct. Clear tag definitions matter, and changed questions or model versions may require checking and labelling products again.
Key takeaways
- Preserve the original records when adding inferred labels.
- Store the model, date, and confidence with each judgment.
- Use database rules, not AI, for prices and arithmetic.