Show HN: Wikipedia and Llama 3 70B

Hey everyone!

I built a little app for exploring Wikipedia.

https://atlaslib.com/wikipedia

You can tap on a suggestion or ask anything.

Why?

Wikipedia is awesome but search on Wikipedia unfortunately is not. I often use Google to search Wikipedia but I wondered if there is a more direct way.

Some of the huge Wikipedia articles (e.g. animals, biology, history, health, science) are thorough but a bit dry.

A language model like llama 3 70b helps with both: discovery and with making a dry topic more approachable.

behind the scenes:

  - client: 150 lines vanilla javascript, tailwind for ui
  - server: flask on fly.io (shared-1x-cpu@256MB)
  - inference: llama 3 70b at 300+ tokens/s via groq

I've set it as my new default page to load when I open a new browser tab.

video (in real-time): https://x.com/__tosh/status/1793699299756228976

I also applied with it for the upcoming YC batch but it is a bit of a last minute pivot from an earlier idea.

Curious what you think and how I could make it better!