Hi there! I made a project for pure joy called Indieseas.net. I made it because I disliked how search engines completely ignore smaller websites, prioritizing ads and businesses. That said, I decided to make my very own search engine! But, how could I keep business websites outside the system? The solution were 88x31 buttons. These are images or gifs that were used back in the days to connect websites between eachother. This was useful because search engines weren't very bad in the times. Thankfully, people still use them a lot nowadays! So, I whipped up a program in Typescript using Bun that used Cloudflare Workers to scrape pages, and then extract links, text, meta tags (title and description) and the images, which I later processed to verify they were actual 88x31 buttons. Now, this setup (multiple bun workers) was super laggy and haggled a lot of resources, so naturellement I decided to rewrite it in... RUST! Now, don't get me wrong, this was way harder than expected, but nonetheless I got a pretty decent result. Now, indexing. It was prolly the easiest part of them all, since I integrated it directly within the scraper and all I did was "docker build" and code a mini embedding API using model2vec/potion-base-32M. I then spun up an instance of the same Docker container and voilá, I've got myself an embedding api that's fairly quick. The last step was querying, for which I made an api directly in the Astro frontend. It's a single SQL query and it works really, really badly. I got to fix it! But the "discover a website" and the "button gallery" functions were very well received by peers and friends.

That's all! Thanks for coming to my mini TED Talk.