A new language for an online store with 10,000+ products.
A big online shop with more than 10,000 products, hundreds of categories and subcategories, and a blog with hundreds of articles ran in one language. To enter a new market, it had to add a new one.
We built an isolated translation service for that specific task: OpenAI-assisted, with Jev checkpoints for quality and relevance. It is semi-automated, and the store's content managers control and observe every step in the admin.
- Context
- Client work
- Catalogue
- 10,000+ products, hundreds of categories and subcategories
- Blog
- Hundreds of articles
- Mode
- Semi-automated, controlled by content managers
- Stack
- OpenAI
- TypeSafe Jev
The task: one language on the site, and the need to add a new one.
A big online shop: more than 10,000 products, hundreds of categories and subcategories, complex templating of SEO titles and meta descriptions, SEO-optimised product and category texts, specific parameters on category and product level, and a blog with hundreds of articles. All of it in one language. A new market meant a new language with the same SEO quality, and the manual effort of preparing product content would double with it.
An isolated service between the store and the model
-
Online store
Products, categories and subcategories, parameters, SEO title and description templates, product and category texts, blog articles.
-
Translation service
Isolated from the store and built for this one task. Translates an article, a product, or a whole category with all its related data.
Jev, the checkpoints
After the model translates, Jev checks the quality and relevance of the translation before a content manager sees it.
-
OpenAI
The model behind the translations: SEO titles and descriptions, product and category texts, parameters, articles.
-
Admin
Content managers check and control translations on product level and trigger automated translation for a whole category.
One category, translated end to end
01 Trigger the category
A content manager starts the run for one category from the admin. The service collects all related data: products, parameters, SEO templates, texts.
02 Translate with OpenAI
SEO titles and meta descriptions, SEO-optimised product and category texts, and parameters on category and product level are translated into the new language.
03 Jev checkpoints
Every translation passes Jev checkpoints for quality and relevance. What looks wrong is flagged for a person instead of going live.
04 Control in the admin
Content managers check each translation on product level, correct it where needed, and approve it. The process is automated, but never unobserved.
Why a dedicated service
Manually: too long
Translating thousands of products, categories and articles by hand and filling them in through a conventional admin UX would be a very long process.
ChatGPT or a plugin: still long, and inconsistent
A simple ChatGPT-assisted approach is still long and inconsistent: someone still has to copy, paste and do all the monkey work, item by item. A generic translation plugin doesn't know the store's SEO templates and parameters.
Isolated, built for one task
The service runs apart from the store, so the store keeps its platform and admin. It knows how products, categories, parameters and SEO templates relate, so one trigger translates a whole category.
Checked, then controlled
OpenAI does the volume, Jev checks quality and relevance, and content managers keep the final word in the admin.
The blog, too
Hundreds of blog articles go through the same isolated service and the same checkpoints.
More use cases
Have a process an agent could take over? Bring it to us.
Talk to UsWe investigate it first, then build the system and stay with it.