Localizer Project

In the past two lessons, you’ve worked on a localizer app that translates app strings into Spanish. So far, it only translates a few words. That wouldn’t be extremely useful for a real app. Today, you’ll get down to business and make your AI agent useful.

Architecture Design

When designing an agent system, thinking of the steps a human would follow is helpful. For a translator, that would include the following:

1. Read the original strings.
2. Understand the context by looking at a screenshot.
3. Take a first pass at translating the strings.
4. Check the translation for accuracy.
5. Improve the translation if needed.

This corresponds to the following multi-agent graph:

Here’s what each part is:

• Input: A text file with the strings and a screenshot image.
• Contextualizer: Adds context comments to each of the strings.
• Translator: Translates the strings into Spanish.
• Checker: Checks the translation for accuracy.
• Formatter: Puts the translation in the desired format.

Here’s the screenshot from the input:

And here are the app strings:

title: Question
message: Would you like to save the file?
negative_button: Discard
positive_button: Save

That’s enough overview for now. You can learn other details as you get to them in the demo.