Bluesky Inline Translator

• 

  BlueSkyInlineTranslaterProjectCover

Inspiration

In today's world, many free online translation services offer high performance, creating an expectation that translation should be free. However, these services often become paid once usage exceeds certain limits, with prices that can be quite high. By offering a limited amount of free usage, these services deter new services from entering the market and hinder the development of applications, effectively reinforcing an invisible language barrier.

To break this status quo, I started working on improving the translation capabilities of LLMs about two years ago, with the goal of running these models locally on my own device.

Now that I have created some LLMs for high-precision translation, the next step is to create a practical translation application for the increasingly popular Bluesky.

What it does

Bluesky Inline Translator provides seamless, real-time translation for both reading posts and posting content in different languages. Unlike other SNS's translation feature, which only allows translation when reading others' posts, this tool enables users to translate their own posts before sharing them, promoting genuine two-way communication.

How we built it

This project leverages the built-in LLM integrated into Chrome, eliminating the need for external API calls or third-party translation services. We used Chrome's LLM's existing translation capabilities without any additional tweaks, instead focusing on seamless integration with Bluesky's user interface. The extension was developed using JavaScript for browser compatibility. It is also possible to use external APIs for languages ​​not supported by the built-in LLM.

Challenges we ran into

One of the biggest challenges was ensuring that the translation quality remained consistent while running locally, without relying on cloud services. We closely monitored the translation quality, particularly for short and relatively simple sentences typical of social media. We confirmed that the default model provided sufficiently high-quality translations for these use cases, allowing us to proceed without additional fine-tuning.

Accomplishments that we're proud of

We are proud of successfully creating a tool that enables true two-way communication on Bluesky, making it more inclusive for users from different language backgrounds. One notable achievement is our decision to focus on a specific, concrete use case rather than attempting to build a broadly general-purpose tool. While Bluesky's user base is still around 24 million globally, which may not seem ambitious or cutting-edge, we believe that by promoting real-world, two-way communication, we can counter the spread of one-sided fake news and make a meaningful impact.

What we learned

Through this project, we learned about the complexities of integrating machine translation into social media platforms, especially the importance of providing a seamless user experience. We also gained a deeper understanding of optimizing LLMs for specific use cases, such as social media interactions.

What's next for Bluesky Inline Translator

In the future, we plan to expand the number of supported languages, further improve the accuracy of translations, and enable voice translation. We also plan to expand this function to specific communities, such as other social media platforms, in order to realize real-time two-way communication with more communities.

Built With

• gemini
• javascript