Backscroll is an agent that scrolls your feeds. It scrapes my twitter and instagrams feeds and imports as much historical likes and bookmarks to send to a llm to compile a feed that you really want to see.
The algorithm I've made for myself is reminiscent of Thunder and Pheonix and includes a mixer that is based on a dozen rubrics that every tweet/post gets scored against. This rubric I set initially and then evolved over time as I've labeled tweets and posts that I like or dislike with the option to also add commentary. Luckily LLMs have gotten much cheaper and a lot more intelligent so even if its scrolling both twitter and instagram feeds at 0.15mph 24/7 it would cost less than $50/week.
This is a in-progress project and I'm still trying to figure out how it works. So far I use it 40m+ a day and it has saved me a lot of time.
I'm working on a denser explanation of this project. Stay tuned!