Brief about weaver
Simply put, weaver is a tool I made for extracting saved acticles from pocket, with a specific tag, and summarize them.
Why building this tool
I’ve been using Pocket since 2014, and I’ve sticked with it eversince. As someone that tends to read a lot of articles at times the Pocket reading list becomes too big to handle, so I decided to do something about it. And because I’ve been tinkering heavily with LLMs (Large Language Models), specifically Anthropic’s claude 3, and OpenAI, basically the idea was as organic as it can be. I just connected 2 dots (The necessity of a simple and fast way to extract articles from Pocket, and text completions/summarizations) and it worked.
How it works?
Basically, one needs to create a Tag and then use Weaver to extract Tagged Articles via API requests. Then, Weaver will use a clever technique which works as follows:
- Weaver will print to stdout a general “metadata” about the article, like an excerpt, the title, url, etc. and will ask us whether or not we want to “save” it (To process it with LLM).
- Then, when the list gets exhausted, it will extract every selected article, summarize it, and save it to a buffer, so it can:
- Join the buffer (list) of summaries, and make one big summary of everything.
- Then, it will print to stdout the “global” summary.
It is like a TL;DR of some sort. I use my own prompt engineering to make it work the way I need it to, but there are some examples in the Github repo.
Ideas: What I do with this information?
With a general summary, I read it in no more than 5 minutes, and if I need more information, I can obviously consult the original article to find out more. It’s an extremely interesting idea to “consume” information faster and more effectively.