It scans Hacker News and a range of subreddits for the most popular tech posts, tags and categorizes them by topic, and generates a TLDR of the important parts of each article. It also runs a sentiment analysis over the comment threads - which I find just as important as the article itself. The collective hive mind of comments is pretty good at rating how trustworthy a piece is and filling in the context the article leaves out, so the discussion summary often tells you as much as the write-up.
The idea is that when you've fallen behind, you can jump into a topic you care about, skim the TLDR and the comment sentiment for each story, and quickly work out which articles are actually worth reading in full.
The aggregation is run multiple times a day. Summaries and sentiment are LLM-generated so they're occasionally off - everything links back to the original article and comments thread, so the source is always one click away. Would love feedback, especially on where the summaries are off of if there's any other analysis that would be useful to surface.
It's all OSS including the LLM aggregation scripts if anyone's interested at doing something similar:
https://github.com/NetCoreApps/techstacks.io/tree/main/scrip...