Kevin Rose

"So first thing I want to show you is that there's going to be some errors here because it's been a minute since I have actually done the whole actually touch this codebase. But what you see here is I have 63 sources of information coming in. And so these are from just your classic RSS feeds. Yes, RSS still does exist."

"And so we're pulling in and saving all of this in a PostgreSQL database."

"So it comes in via RSS I hit Iframely and Iframely gives me back additional metadata about it so like title description sometimes it can pull in even additional kind of deeper context about it article data things like that and I want to store all of that that's really important for me to have that rich data."

"And then I can see that Gemini has been run against it as well which we can get into for as to the reason why. And the winner has been resolved, the TLDR has been generated and the embedding has been done. So the winner is basically saying, okay, there are a bunch of different factors here that might come in."

"And so you can consider Firecrawl to be a very similar thing. It is a little bit more on the crawling side. So, a little bit more fine-tuned around crawling. They have some AI aspects to it as well where AI will actually try and go out and figure out how the best way to kind of scrape the content."

"And then what I hit is I hit GPT-4 Mini largely because it's fast as hell and cheap and it's very it's quite smart."

"You can create mathematical representations of that content and embed these with OpenAI and store them in PostgreSQL by using a vector extensions."

"And so what I do for that, actually, let me just do a screen share again, is for the durability side, I use a service called trigger.dev. And so I like trigger.dev because what it does is it allows me to create these functions and they're all TypeScript that live in the cloud."

"And then will report back to me via Sentry or any other type of monitoring software that I have on the back end that oh I had a failure why did this actually happen right."

"I use a Tavily search API and you can say go out and find me other stories outside of my RSS ecosystem that may match this and bring them in."

"What I would do is I would just you know use compound engineering on Cursor and do a workflow."