Another Dumb App?

Yes.

It’s been a while since I’ve posted, but I have been toying around with stuff. One of those things is an app idea I’ve had for a while. It addresses one of the biggest issues affecting us today—being stuck with an empty beer at a bar!

This sounds like the only legitimate use of computer vision that I can think of.

Explain The Name

Where did brrrvay come from? As I mentioned in an earlier post (that I won’t link to), every name is taken. So I took Beer Ray Vision + Money Printer Go BRRRRR and somehow that became brrrvay. Pronounced burr-vay.

How Does It Work?

Whoa, who said it works???

It kind of works right now. I may do another post on the architecture, but short story: brrrvay has a lightweight agent (Python) that runs locally on a small computer, connects to USB cameras, and uses a Roboflow workflow (or a custom object detection model) to identify empty glasses in real-time. The agent reports events to a cloud-based server that supports multiple customer accounts, provides a live admin dashboard, and manages alert logic.

What’s This Post About Then?

The grand vision is to allow multiple types of alerts when empties are detected (e.g., push notifications, flashing lights). I want the web dashboard to have a funny alert built into it as well. There was a money printer go brrrr site at https://brrr.money (it was down when I just tried, maybe that’s temporary) that I thought was kind of cool. It basically let you ramp up the chaos of a money printer.

Looks like you can still get a taste here https://github.com/memetic-institute/Money-printer-go-BRRR

Using that as inspiration, I ended up using a couple of images with varying levels of jittering based on configurable empty beer count.

You Going To Finish This Or Nah?

I doubt it. I would like to see it get to a state where it works pretty well for me. The main issue right now is my model needs to be trained with a lot more annotated images of glasses with varying levels of emptiness, along with other chaos that will be present at bars (e.g., cans, bottles, handguns).

I’m not above more field testing though. Evidence below.

That model accuracy shouldn’t really be that big of a deal. It’s a problem that could be solved with a little elbow grease, and that could get it to a usable state. The bigger issue with making it a platform for bars and restaurants in general is a hardware issue. Figuring out what types of edge devices we’d allow (or sell). A lot of options, but mainly deciding if people can bring their own compute and cameras (😔), or get the Hooli Brrrvay Box Signature Edition (like a Jetson Nano with our approved cameras).

brb, I have to get back to field testing.