Frey Chu

A1 containers, there's a two stall trailer and a four stall trailer.

Again, this is gonna take more time, but you can set up whatever logic you want.

And actually there's a lot of nuance with this process.

So just to kind of go over that, like there's a ton of edge cases that you wanna check after every single pass.

And one of the first mistakes that I did when I started enriching my data with Crawl 4 AI was I would just give it this massive laundry list of things to get.

and just gave it like a massive csv and i was like get the trailer inventory get the images get any amenities features pricing all of that stuff and it just didn't work like it was super low quality um so that's why i'm kind of going slowly one at a time and this is also helpful because after this pass i could examine the results

You always find edge cases, these one-off situations where you should let Cloud Code know that it messed up, it done messed up, so that it can fix it.

And you might have to rerun this.

I reran this, I think, two or three times until pretty much all of my data was good and I had all of the stall information for every single listing.

So yeah, this process just continues one by one.

For me, the next one was trailer images.

And this one was cool because not only did I use Crawl for AI to scrape the images from the business websites, which by the way, I know is a gray area, but I plan to go and reach out to these businesses and ask for permission and just have them claim the listing, which basically gives us the green light to use this.

But

You can scrape the images and what I did was I just told Cloud Code to look for any alt text file names, examine the page that it's on and scrape the highest quality images.

And then what I did is I told it to create, I told it to take the top three candidates for trailer images and then I just sent it over to Cloud Vision.

And this way it was able to look at these images

and identify the best ones.

Because the first time I did this, I got logos and crappy images and favicons, and it was like, am I really about to clean this image data?

This sounds so bad.

And this is the workaround that I came up with.