Examples
Feature tour: real-scenario examples
Designed around the task, not the feature. Every rule is validated against a Before/After simulation before it reaches you.
Prepend brand only when it’s missing
Keep titles consistent without creating duplicate brand names.
| Brand | Title (before) | Product type | Title (after) |
|---|---|---|---|
| Nike | Air Max 270 | Shoes | Nike Air Max 270 |
| Nike | Nike Air Max 95 | Shoes | Nike Air Max 95 |
| Adidas | Ultraboost | Shoes | Adidas Ultraboost |
Normalize legacy naming
Replace an old naming convention in bulk so historical data stays consistent.
| Title (before) | Title (after) |
|---|---|
| Women's Fleece Jacket | Womens Fleece Jacket |
| Women Running Shoes | Womens Running Shoes |
| Men Leather Belt | Men Leather Belt |
Drop dead source values
Clear placeholder values before they reach downstream, so "N/A" never shows up on a product page.
| Brand (before) | Brand (after) |
|---|---|
| N/A | (cleared) |
| Acme | Acme |
| (left empty) |
Tag seasonal items
When product_type matches a seasonal word, append the tag to the end of the title.
| Product type | Title (before) | Title (after) |
|---|---|---|
| Winter Jackets | Northstar Trail Jacket | Northstar Trail Jacket · Winter Collection |
| Shoes | City Runner | City Runner |
Related questions
Where can I try these examples directly?
The rule builder on the homepage (#workbench) ships with sample data and example prompts. Click "Load sample data" to see the Before/After simulation right away.
Are the rules in these examples real LLM output, or deterministic?
In the MVP, intent parsing is deterministic: supported operation patterns (prepend / replace / clear) map straight to a structured rule and are validated in local simulation, so a guess is never treated as a fact.