Hacker News Reader

My guess is the roast version looks at the repo details/descriptions (or at least a few of them - which is why the reply seems much more specific/cutting) where as far as I can tell the praise version here only looks at readme/top level profile info.

I think the reason here is because one is using Llama and the other is using GPT-4o.

But as another commenter pointed out, they don't seem to consider your repositories but simply your profile

I could see it being a lot funnier if all the praise was sarcastic.