gkamer8

I got a bit carried away trying to make a WiFi captive portal. A lot of the tutorials online are incomplete or underspecified, so I wanted to make sure this gets indexed.

Lmao really fun. I guess you're using a network request, but I wonder if you can access the models that the photos app seems to use locally.

Yes, sort of - that and scaling reasons. It's actually in that same repo now but in a different service. I'd like to remove it from the Abbey repo entirely eventually.

Oh wow, totally forgot. Just added MIT.

Hi, it is unfortunately not public and cannot be made so to my understanding. It was frustrating to talk about in job interviews for that reason and therefore was not on the LinkedIn.

Hi thanks! The domain actually used to be a redirect link to U.S. Automotive Industries (a trade publication). I reached out to them and got a deal, so it was a lot for me but not, like, we.ai expensive lol.

The name was always a corporate placeholder and I liked the idea of US Steel or General Electric type names. Some startups have done similar things, and many people actually like the name a ton. But I know it's controversial and so any products I made have their own names and branding that's pretty separate (see: Abbey).

Over the past few months I've gone the gov contracting route and the name actually made some sense, so I've used it raw. Still, the plan is to get a DBA in the near future and switch it up. Thanks for the advice!

Thanks, that's a great idea! I was originally going to go the vision model route because I'd also like people to be able to send instructions to sign in with some credentials (like when visiting the nytimes or something).

Thanks for the catch on my LinkedIn, I really should have that there now. It was originally something I kept private.

I'll try to improve the vibes :(

I've been working at this startup for almost two years now and that page and branding etc has been changing a lot as you can imagine ...

Yeah the EU cookie banners are annoying, I'm hoping to do some automation to click out of them before taking the screenshots

> Almost like it's designed to nerd-snipe someone into a PR to change the format based on Accept headers

pls

We released the Workspace feature 2 weeks before Google released theirs, which was shockingly similar obviously. Watching Google succeed (?) with notebook LM has been... frustrating. (ofc they did not copy Abbey, it was just annoying)

Hey, thank you! Could you explain that point about private collections more?

I've hosted Abbey for sometime at https://abbey.us.ai and now made a version you can self host. Students use it a lot for reading.

On the whole, it orchestrates a variety of models (TTS, OCR, LLM, etc.) and other tools (search engines, file storage) to make a configurable, private AI gateway. You can plug in your own third party API keys or self-host the models as well using Ollama.

I had often gotten frustrated trying to parse the code at a bunch of different layers of abstraction just to make one change to a model or just see one detail of the implementation. On top of that, matching pre-trained weights to those edited models was pretty hard for me.

Agrippa consists first of a Python package that takes model architectures written in XML syntax and compiles them to ONNX, a general neural network format. That file can then be imported into PyTorch using the package and trained. Afterward, the model can be published to the zoo on the website. Other users can import entire models or parts of a model into their project. On the zoo, the XML makes it really easy to visualize the models down to every operation.

Overall the markup ideally forces people to make their code more local, so that all of the details for any part of the model are readily accessible. The details are also more explicit - things like initialization and weight sharing. All of the parameters are explicit and named as well.

Hey, thanks for the feedback! I just went and added the extra "why" component to the Readme. I'll try and add some more when I get out of class.

Also would be happy to hear feedback/fix problems/add features if you'd like to use it at all.

I got sick of scanning through large json files to try and get a feeling for a dataset. Then when I wanted to remove some data, I was using control F - this is obviously dumb. Praetor is a flask app that lets you mix, edit, view, and export datasets for LLMs. The only pip requirement is flask. To keep it simple, the frontend is very straightforward. It uses an sqlite database and jinja templating under the hood.