Transcription
I had Cloud Code create me a Twitter scraping bot, and let me show you how it works.
So, every 30 to 45 minutes, I get a message inside of Telegram that is showing me the most like trending or viral tweet in the AI space within the last hour. It gives me a link to the tweet, what the likes are, how many retweets, when it was last posted, essentially a whole velocity score, and I can actually respond to it inside of Telegram.
And the way it works is actually pretty cool because this was built through Cloud Code. It was pretty easy to do, but it's relatively sophisticated and very customizable. So, every like 45 minutes, it actually calls an Amplify tweet scraper, which is pretty cheap, to grab like 40 to 90 tweets. From there, it filters the tweets and then ranks them. It has essentially five scoring signals. So, velocity, like how many likes and retweets has it gotten since it got posted, authority, aka who's posting it, timing, opportunity, and replyability.
All these tweets get pushed into Superbase to make sure that we don't have the same tweets showing up over and over. And we also get some diversity of authors. From there, it ranks the picks, gives them a probability score via softmax, and chooses one. The tweet is then pushed to me on Telegram, and I have the ability to respond to it via a pre-created response with Grok, or I can do my own.
Now, the nice thing is to kind of get away from like the constant AI slop spam, all these responses are sent to Superbase. And so all my responses essentially get graded based on a score the AI provides and myself. So, over time, we're able to see like which tweets do well, which ones better match my style. And so it kind of becomes like a self-correcting system.
The bot itself is hosted on Railway, and then I have a front-end dashboard hosted on Vercel. So, um, if you want to get your hands on this, it's actually inside of Chase AI Plus.