Localization Hub: every translation request in one place
invented data
A web app for the whole localization process. Teams file proofreading and translation requests, each request is assigned and tracked across every market, and progress from Crowdin comes in live. It began as a one-file prototype and became a real multi-user app with logins, a database, dashboards and reports, running on a cloud server. A machine-assisted approval trial was tested on a dry run before any live use.
Click a request to move it to the next stage.
Taking a tool from a first prototype to a live multi-user web app.
Catching a chatbot's wrong answers
invented data
Every week, an automatic audit compares what a support chatbot told customers with what really happened in the billing system.
Testing AI output against real data, and measuring how often it goes wrong.
Translating help articles for many brands
invented data
A pipeline downloads each article, translates it and uploads it for each brand, keeping the images and checking the links. Pick a language to see one result.
SourceYou can cancel at any time from your account settings.
Building multilingual content pipelines, from source text to published page.
Teaching the AI not to repeat a mistake
example
Errors found in native-speaker review go back into the agent's instructions, so the next batch starts better. The same loop runs in my small handmade textile business.
- AI draft
Attualmente, preferisco il treno.
- Review
"Attualmente" means "currently". It should be: In realtà preferisco il treno.
- New rule
Never translate "actually" as "attualmente".
- Next batch
In realtà preferisco l'aereo.
Improving prompts and agent instructions from real errors.
Auditing 485 reply templates
invented examples
Every ready-made reply the support team uses was compared with current prices and help content. The changes went live in stages, with a dry run and a way to undo them.
Auditing content at scale and rolling out changes safely.
One person, a team of agents
invented notes
One AI agent for each area, coordinated through a shared noticeboard. Each one owns its part of the work.
Rule: never edit code you don't own. Leave a spec for its owner.
- Server agent: Deploy script updated. Dashboards, please test.
- Chatbot audit agent: Need a change in the support tools. Spec left in their folder, code not touched.
- Dashboards agent: Tested live. Done, moved to the recent log.
Running several AI agents on a large body of work without losing control of it.
Dashboards that keep themselves current
invented data
Support tickets and chatbot conversations in one dashboard, refreshed daily. A weekly report runs by itself, and a watchdog rebuilds any week that gets missed.
Share of tickets by topic, %
Turning raw data into reports that run on their own.