Very cool and super clean website! Talk about your infra please!

Thanks!!

Infra: I was hosted on Google Cloud for a while -- literally a single VM running Docker Compose, but I decided I wanted something a bit more flexible and interesting, so last month I switched everything over to Fly.io and I am incredibly happy with them. It's just so easy and fun to manage.

The retail site (rarity7.com) is just a small VM running a Crystal server process to handle web requests. Image hosting is all done on Cloudinary. My backend / inventory management / trading + research engine is a separate Crystal process in a separate VM. Both connect to a Fly Postgres DB. There's one other service which is a small python process on another VM which is doing inference on my regression model. That's super lightweight and I don't need any GPUs to do the inference (tabular data is nice like that).

Overall it's really nothing fancy and it works quite well. A few web-serving VMs and an inference service for my pricing model. I train/retrain the model a few times a year on a local box (my repurposed gaming rig running a 2080Ti).

As @mplewis said, because it uses the SV API - however, I've since found a way to reduce that cost to practically 0. See https://loichovon.com/posts/streetview-scraper-v2.html

As you say though, another way to go about this would be to scrape the public SV site, which I have not attempted to do. @xnx linked this project in another comment, which seems to do just that: https://tllabs.io/google-street-view-hyperlapse-youtube-vide...

I don't know if/how they evade captchas.

Say more? I don't understand the technique youre describing and its objective.

I automated clicking the Streetview arrow in a browser and taking a screenshot of a defined region of the screen to create a type of Streetview Hyperlapse. Many other tools have done it better since, at least as far back as 2013: https://tllabs.io/google-street-view-hyperlapse-youtube-vide...

Ahh I see, those hyperlapses look very cool!!

Going back to your first comment, I originally made this as part of an app with an interactive streetview component. I reused the initialization code I had as the base for this. Re-using that code also makes sure the screenshots taken would be close as what you'd see using the embedded streetview (In our project we had a bunch of remote locations without a lot of SV images, so depending on the search radius used when querying for a panorama, we could end up in vastly different places).

However, I realize that I never even considered scraping the public site, which would allow for hiding your identity by not needing an API key. I'll think about this.

no issues yet. I'll update the post if I get in trouble :D

Yeah but you're limited to a max size of 640x640, get a Google watermarks and it costs .007 USD per image.

If you check out my other blog posts, I use this as a base to build a scraper that's essentially free.

> The goal of the design process is (...) to instead characterize the design space for a given problem

Any good resources on how to properly define the design space? What makes it a space more than a list of requirements?

I am interpreting it as an n-dimensional graph (let's say 2D for simplicity), and the axes may be something like "speed" and "scale", or "cost" and "risk", or whatever your parameters happen to be. The solution is somewhere in the space defined by those axes, but you don't know where it is yet, and you don't want to lock yourself in to a single point too early. The axes themselves could be anything. You are choosing what to care about. Ideally, it should be informed by understanding your users, your stakeholders, your domain, your market, resources, available technology, divine inspiration, etc., but it's an important decision that is probably completely contextual and is part of the design process.

I could be wrong, this is, as noted, just an interpretation.

*coup de grâce ;)

Sorry my comment wasn't very clear about this. There's still secret sauce in the materials. The GDS is purely dimensional. It's just a bunch of shapes. But how those shapes get translated into actual silicon, and how various chemicals are used and in what concentrations to manipulate the silicon to make the transistors function is the secret sauce.

What could be nice is the developer providing some unit tests, and Second using those to ensure the PR is functional, re-generating if not.

Given a set of tests, I guess you could try to generate the code as well, though I could imagine it might overfit on the given tests and create a "hack" solution.

One of the great things about Second is that uses GitHub as the handoff of code between bot to human. This means you can run your own CI against the code changes that the bot generates, triggered by GitHub actions.

I actually find that I'm more awkward online/in chats than IRL.

The fact that you have longer to think about your messages and modify them before publishing, that you have to think carefully so as not to multi-text (which seems desperate), that you don't have access to non-verbal cues for context, can't smile etc. all contribute to this.