Behind

Global Version Website Development with GPT & Copilot

When developing a website, you might focus solely on your native language, but for a SaaS service with global sales potential, something feels lacking. However, aiming for a grand global launch can be more demanding than expected. Our team has been tackling this challenge with our unique DIY approach as outlined below.

User Message Handling

Drawbacks of Variable Processing Method

Typical i18n libraries handle multiple languages as follows: they create a key for each message and display the corresponding message in the user's language. At first glance, this seems like the most logical method.

This method represents the simplest form of running a multilingual version. As the business grows, it has the advantage of allowing flexible addition of various languages. However, when the business hasn't reached a suitable level of growth and the work is being done by a small development team, transitional issues arise, such as:

While these are minor chores, they result in the wastage of already limited development resources when developing the front end that interacts with user messages.

Embedding Language Messages in Code

To minimize these inconveniences, the Videostew team is developing multilingual versions at the speed of a single native language version. Instead, we are limiting the multilingual version to only "English." This approach might not be sustainable as the market expands, but it's extremely useful in the initial global phase for appealing to the "English"-speaking user base.

At VideostewTeam, we register and use a global helper called d(irect)trans on both the backend and frontend. The implementation of this code is quite simple. It takes a Korean message, which is your native language, as the first parameter and a global English message as the second parameter.

The part where messages are exposed to users can be simply created as shown below.

While it may seem like you still have to translate into English when coding, that’s not the case if you use it with Copilot. Copilot, or tools like Cursor AI, reference the helper functions you frequently use or the definitions in open tabs, so they understand the usage of frequently used helpers very well.

As shown below, when you input Korean into dtrans('.., it can infer that the translated English message goes into the second parameter, allowing you to create a global version while coding in the editor.

< Type in Korean, and the auto-translated English is added >

This approach overcomes all the drawbacks of developing a formal global version mentioned above.

As mentioned earlier, this method is aimed at quickly implementing a global version without much effort. It is not suitable for creating more sophisticated translations or multiple language versions. As the service grows, the language system should be changed.

Automatic Translation of Guides & Blogs

Besides user messages in code, our service utilizes a multitude of messages and texts. Fixed pages like landing pages or terms can be translated once and that's it, but for dynamically created and updated content like guides or blogs, manually keeping track can be quite a hassle.

In response, at Videostew, we register posts in our native Korean and then, after a certain period, estimate the modified parts and use a GPT model on the backend to translate them into English automatically. While there are various API services for translation, using LLM for translation offers the following advantages:

When it comes to practical application, it's impossible to translate entire texts each time for large guides (almost book-length). Therefore, we generate hash keys based on content per paragraph and track only the modified paragraphs for automatic translation.

< Hash values for modifying only the changed paragraphs >

This type of call (which we are not currently using) could potentially allow for cheaper bulk translations using GPT's bulk call discounts.

Customer Support Response

To provide a perfect service experience, we don't use external chatbots but have our own. It's not exactly a chatbot; we use a modified version of our real-time comment-capable bulletin board. If a user's information is not in their native language, it automatically attaches the original text in their native language. When we reply in our native language, it's translated back into the user's language and delivered.

< Questions in languages other than your native tongue? We've got automatic translations for that! >

This reduces delays caused by translations and allows us to provide seamless global services.

Wrapping Up

Many services are already integrating GPT into their development and operations. For small teams like Videostew aiming for global reach, I hope this blog provides a few helpful hints.