• 0 Posts
  • 10 Comments
Joined 1 year ago
cake
Cake day: June 30th, 2023

help-circle


  • A school district spends $180,000 (hyperbole, I don’t know actual numbers) of taxpayer money deploying this system between the actual hardware costs, maintenance costs to install the hardware, it costs to implement it into their network, and probably an ongoing contact with this dummy’s company. Maybe only for support but with the way things are now I’m sure they built this app to phone home to their servers (introducing a huge potential security risk over simply running it locally on the schools existing network infrastructure in a docker or something), calling it “cloud based”, and charging the district 1k/month to run the devices the district now owns and should be able to operate without the company. The company then talks about how they’ll back up records and safeguard data so you don’t have to worry about that (that it dept you pay is pointless!)

    Three months after deployment it turns out the sensors can be tripped by many things not related to vaping, maybe increases in heat, mouthwash breath, etc. the false positives are due to a hardware flaw and cannot be fixed with a patch. Feel free to upgrade to sensor version 2.0, now with improved accuracy! (read: the problem still exists but isn’t as bad). Only another 40k to buy the new hardware, rip out the old hardware (which is now worthless), install the new stuff, and configure the software for everything (again, maintenance and IT costs)

    9 months after deployment the company is doing poorly because their product is stupid and only a few idiots actually bought it (way to go idiot). There’s concerns because they sent a new Eula that outlines data sharing policies. They are potentially finding ways to harvest the data they agreed to safely store to try and create a new revenue stream to right their sinking ship. District counsel says fighting the Eula change will be expensive and there’s not much precedent for it, plus they state they will anonymize data before sharing so it’s not a ferpa violation, technically. It feels scummy but you can’t do anything about it. You also don’t really trust them to only sell anonymized data but you can’t prove they aren’t crossing that line so whatever, I guess

    15 months after deployment they get hacked because they’ve run out of vc cash, never could get an actual profit stream going (turns out they’re spending 750,000/yr on salaries for 5 people and they’re all kitted out with sick work computers for what is basically coding a web app, but I digress). security of their servers was one of the budgetary constraints they chose to make to right the ship (but had to keep the $1800 office chairs and the 15-20k/mo rent loft they use as an office in a hcol area). The contract says this may happen and they’re not responsible unless there’s gross negligence on their part, which you can’t prove, and that they do some bare minimum reactionary shit after the fact to mitigate damage. So they’re legally blameless and now you get to notify your community their children’s data was leaked to god knows who, whoops

    22 months after the fact they go out of business officially. You get a form email about the company’s journey and the difficult decision they had to make to stop fucking around on a dumb project that sucks because no dumbass vc will give them fun bucks anymore to keep playing tech bro billionaire. All the sensors stop working because they require a connection to the servers, which they shut off immediately without a sunset period. You’re reminded every day when you log in to the schools admin panel and get 350 “sensor not connected” error messages and your students bitch about the “sensor not connected: server not available” error pop up showing up on their classroom console. It takes IT a few days to remove their shit from the network and that costs you even more money in wasting your IT staff time when they should be fixing the broken computers in the computer lab or whatever.

    Now your school has a bunch of weird boxes on the wall. Sometimes people ask you about them and you go “oh those don’t do anything” and remember that they cost taxpayers in your community tens, if not hundreds, of thousands of dollars and wasted hundreds of hours of your supports staffs time that they could’ve been using to improve the school

    But then you scroll on instagram and see there’s this new thing that will detect when kids are bullying each other. You just have to put a camera in each classroom. It’s okay, it won’t record. It will just use the power of AI and machine learning. You’re sold right there and the cycle starts again



  • Do you rent or own your place? If you have the ability to run Ethernet drops to camera points imo the best solution by far is to get poe cameras. Amcrest/reolink are good options. You’ll need a poe switch, this can be expensive but you can pick them up cheap used if you look for old ones sold as ewaste. I got mine for $40 and it has 24 ports which is far more cameras than I’ll ever need.

    Similarly, you can get a nvr, which is basically a poe switch with a built in management system and a slot for a hard drive. These can be a couple hundred. Or you can get an old ewaste pc, like literally an $50-$100 sff core i3 pc from an office off ebay. Thow a decent sized hard drive in this, the bigger it is the more recording you can do. I have this connected to my home server for storage so if you have a home server/NAS that’s also an option but not necessary, just gives you more record time and eliminates the need for buying a computer to act as a server.

    Then software to tie it all together: ZoneMinder Moonfire NVR Frigate MotionEyeOS OS-NVR Are all good options Also closed source options like shinobi, I spy, blueiris, and a ton of others

    Not revealing my specific setup for opsec

    This imo is the best possible setup. For one, it sounds expensive but ultimately costs just a bit more. You can get super expensive poe cameras but comparable poe cameras to wyze/eufy/etc are often a bit cheaper because they don’t have the WiFi nonsense built in. Of course, you pay that back with the switch and server.

    But the bigger thing is reliability and customization. Before this I had a eufy cam setup. They were wireless which was admittedly easier to setup, no fishing wires. But every couple days I’d get notifications “camera x is unavailable” for no reason. My home has a very solid mesh WiFi network with several APs. The cameras are just shit and drop connection randomly. Sometimes they’ll be on for 4 weeks straight, sometimes they’ll disconnect 20 times an day. If you have a setup with 10 cameras it means one is always doing it.

    Then eufy came out and was server siding thumbnail id images, despite claiming to never do this. Then they doubled down on this, and took away the guarantee that they wouldn’t “cloud” your shit. Essentially they would do “ai” facial recognition server side because their little base stations aren’t powerful enough. They’d then store thumbnails of recognized users for future id purposes. This caused me to sell the eufy cameras and go poe. The poe cameras work in an isolated vlan, eg the cameras and all their features work without a connection to the internet and I can tunnel to my server to view them remotely. You don’t need to have this setup but I’d recommend it if you can

    Finally going off the above with your own server and your own hardware you can do whatever. Eufy had ai recognition but it was shitty. I’m sure it’s improved a bit. I’ve found running the models locally appears to be better, more features like yard perimeters, object detection, etc. you can also separate the ai model from the nvr software, etc. frigate is an interesting potential here, still needs some growth wrt object detection but if they get it a bit more mature imo will be a serious contender


  • Flac 44.1 16bit level 3. Host with something that meets your needs. I have my files in jellyfin and navidrome and can then access the library remotely either through jellyfin web client, navidrome web client, substreamer, Finamp, kodi, etc. but this way if another amazing format comes up down the line I will always have my library in a good state to transcode from. Tag and sort everything with beets.io (or musicbrainz picard is great, I just like that beets is cli). This results in a library I can access on my phone, laptop, tv, carplay, etc

    Technically you could go for 24bit but imo the extra file size isn’t justified. though one could make that argument for flac vs 320cbr mp3, transcoding 320 mp3 is more likely to create artifacts, thus the reason for keeping around flac

    Alac may be easier for you if you use mac


  • Or if you have an old machine and a enough money to by a few hard drives (which you should if you can afford a synology) throw the drives in the old machine and slap something on there. Truenas, Proxmox, unraid, etc. unraids probably the easiest but it costs money. All of them have some kind of docker/kubernetes so you can just run whatever open source version of the thing you want. Nextcloud, libreoffice, etc. you could just install some version of linux too, doesn’t need to be one of those, but those are much simpler to deploy and (most of them) are tailor made for the task

    Synology can do all of this too but isn’t as expandable. Want more power to run a jellyfin server and transcode 8 4k streams at once? Plop in a gpu or better yet upgrade to an intel with quicksync for low power usage. Want 8 more hard drives? Change the case and add an hba. Want 24 more? Add another hba and a disk shelf, as long as your motherboard has enough pci lanes. It doesn’t? Upgrade it. The trade off is usability, the synology stuff is easier to use. It’s also more expensive initially, you can make a basic nas with a $50 e waste pc that an office was throwing away (though tbf you’ll probably spend a bit adding disks to it just like you would with a synology)

    Depends on how much of a dork you are I guess