Skip to content

Link Roundup - Pulumi, ActivityPub, Neural Nets

esmevane's 'e'
Published: at 12:00 AM
4 min read

Here’s the tabs I hung on to for much too long. I kept them around because I could imagine reading them, but I haven’t spent the time. Time to declare bankruptcy!

ActivityPub Test Runner


Bengo and Bumblefudge, two folks who actively participate in the ActivityPub ecosystem, have been working on a test runner for Activity Pub implementations. I think I’d love to reference this when I start putting ActivityPub support into my Rust side project.

Pulumi Automation API


Speaking of my Rust side project, I’ve managed to get a bunch of stuff put together for its infrastructure over the last few months of spare time. I can now provision a rough prototype of the platform and then start deploying to the platform from Pulumi.

What bugs me about it, though, is Pulumi runs all these stacks and setups on its own. Enter the Automation API! This can let me set up a whole new service that just works as part of my base platform, which I can call to manage things.

Here’s an example of a Typescript node running a stack deployment as a CRUD interface.

I could imagine throwing up a node app in my orchestration setup phase of provisioning and then having that node app be in charge of managing stacks. I’d like to know more about streaming its responses and stuff like that, so it can be as realtime as possible.

No rust support

The other thing that bugs me is that this all needs to be run in Typescript, Go, or other blessed languages from Pulumi. That’s pretty close to a deal-breaker, I think. I want my stack to be primarily Rust on the command line and systems level, so that I can ship binaries that I can trust and won’t have to tend to much after I draft them.

There’s a discussion about it here which I’m keeping an eye on, but it isn’t promising so far.



Vector is an observability platform written in rust, which supposedly manages the entire thing from collection to aggregation to dashboards on the other end. I’m considering it for the side project right now, since I would like to have more or less total control over every deployment other than some key things at the boundaries, like mail and SMS.

Machine Learning Resources

Like it or not, the industry is adopting these things and it’s time to learn them. I’ve followed this space for years now and I’ve mostly found that the tools out there are promising. Although they’re overhyped in a lot of ways (SERIOUSLY overhyped in a lot of ways), the nature of classification and tiny generative assistants is probably going to factor hugely into software going forward.

Here’s some learning sites I found that feel interesting and robust.

Practical Deep Learning


This is a free course that shows you, like it says, a practical way to get into deep learning. This course is supposedly around 8 hours of length, all told, so I imagine it as something which would take a few weeks to really get through and digest for a new learner.

Neural Networks: Zero to Hero


This is by Andrej Karpathy, who is a huge evangelist and educator in the machine learning space. Andrej just recently left OpenAI at the time of writing, amidst a lot of bizarre and weird goings on, and has begun posting cool videos on Youtube as well.

I don’t know about you but “leaves OpenAI, immediately starts sharing knowledge being hoarded for profit by OpenAI” is good enough for me in terms of vouching for this guy’s character. I want to hear what he has to say.


That’s it for today! Thank heavens it wasn’t a big one. And would you just look at all that fresh tab real estate? What a relief.