Nebulink

Nebulink has been optimized for different form factors and device specs. Here it is running on an Android phone.
Here's the standard home page on a 16" laptop screen.
Here's the standard home page on a 32" desktop screen.
You can even run the client on a watch! AI features are limited to translation via server.
You can also run Nebulink and its AI summaries in VR mode on headsets like the new Galaxy XR

Inspiration

I use Mastodon to read developer updates and get the pulse of tech-focused individuals.

Although I appreciate the open nature of the platform and the lack of large corporation developed algorithms, it would be nice to have some features which are readily available on similar platforms (Reddit, X, etc.).

To start, I thought quickly getting consensuses and summaries would be a cool feature.

What it does

Nebulink allows you to read your Mastodon feeds and summarize the contents of "threads" and comments.

You can also run a quick sentiment analysis of your followed hashtags or of the popular feed. The AI features are meant to enhance your viewing experience without completely taking over.

Because Mastodon has limited recommendations, I thought a rudimentary recommendation system might help new users to curate their experience. Share what topics you like reading about, and Nebulink will search the popular feed to see if there's anyone posting about them.

How we built it

We used as much vanilla tech as we could. Instead of a JS framework, we're running pure HTML, CSS, and JS. The backend is a barebones go server.

For AI APIs, we used Summarizer (to summarize posts and comments), Language Detector and Translate (detect which posts need translation then execute), and Prompt for sentiment analysis and classification.

When necessary to get a little more power out of the web platform, we make use of Babylon JS (for 3D and WebGPU/WebXR), and Wllama (for fallback AI implementation), and a custom translation backend for server-side translation when necessary.

Since we work, we used our weekends to put this together.

Challenges we ran into

Unfortunately, the AI apis are not available on all platforms. From the beginning, this app was intended to be a cross-platform Mastodon client.

The web is what we reach for when we're looking for easy cross platform compatibility, and I that was an obstacle.

Accomplishments that we're proud of

It's largely a vanilla experience.
It works across devices.
The Samsung Galaxy ecosystem can use this app and expect the same (relative) functionality. That includes on your Galaxy Watch!
## What we learned
This set of APIs would be much better if it were at least available on Android. Without access to these APIs, their functionality is somewhat limited.
I also learned how useful these can be if they're not behind a flag. I'll certainly use these in more serious projects for text searching and name matching.

What's next for Nebulink
The recommendation system should get more polished so that new users don't have to think about which server to join and who to follow.
More features for posting and interacting with the Mastodon platform. I'd love to further adapt the XR functionality, too.
We'll add support for threads, bluesky, and possibly some other fediverse applications over time.