I'm rewriting the open-source core [0] of our product [1] to add more flexible reports.
This is the first time I'm building with the help of AI (Claude), but I don't really let it write code. It has been really helpful in discussing ideas (I'm mostly working alone), like a rubber duck that actually talks back. The code it generates however has many subtle and sometimes obvious mistakes in them, so I decided to write the code myself.
The main benefits over the previous version will be:
- Dynamic reports (ClickHouse queries are generated on the fly)
- Better performance
- Much cleaner and less code
- Better and faster data ingestion including new features, like full support for JSON metadata
- And much more
It feels kinda crazy that I've been working on this project for more than 5 years now and it's still exciting :)
Of course not, otherwise the rich wouldn't get richer and that would be bad for power consolidation. Everything is moving into this direction, so why should it be different with AI?
We had this discussions for years with factory machines, and nothing came out of it. Don't get me wrong, I hate having taxes for everything (living in Germany with ~65% total tax strain for me if you include everything), but this is about power and stealing other peoples work.
I've decided to take a break from working on Pirsch Analytics (pirsch.io) until next year and focus on some side projects instead:
1. Shifu (https://github.com/emvi/shifu) - a code-based CMS with admin UI. It's really easy to set up, written in Go, free and open-source, and I already sold a few websites using it. It can be used as kind of a framework to build more specialized features into a website and takes away the maintenance hell from managing a WordPress installation or a similiar CMS with tons of plugins that break with every update.
2. Zenko (working title, repo is private for now) - a very simple and no-bullshit project management software. It will be free and open-source, but I might offer a hosted option for a few bucks (like $20/year for all users of a team). I mainly build this for ourself to replace Linear, because we don't really make use of it. Don't get me wrong, Linear is awesome, but we basically only need an advanced Todo list. Main goals:
* Pull updates on the dashboard by yourself, instead of receiving notifications all the time via email
* Keep it simple stupid - no unnecessary features, no AI, just the bare minimum
* Cheap (for the hosted version, free if self-hosted) and easy to host (again written in Go)
* No feature-creep
3. Last but not least, I'm working on a "game engine" written in Go and SDL2. I do this for fun, but it is coming along nicely and teached me a few new concepts already (like ECS in Go).
On the plus side, this has prompted many people to finally switch to Linux. Even people I would never have thought would consider it are now thinking about it, or have already moved over. Companies are also recognising the issues with Microsoft.
HashiCorp Nomad is a nice alternative is something you can look into if you're looking for a lightweight Kubernetes replacement. Also, it doesn't support everything Kubernetes does, it should be enough for most use cases.
I was shocked how well even tripple A games work out of the box when I installed Metro Exodus recently. I get 60fps+ on 4k at max settings on my 1080TI on Manjaro Gnome. Here is a screenshot: https://twitter.com/m5blum/status/1312878889711079426
I still keep Windows around as a dual boot option for games that don't work well with Proton though, but I barely need to use it.
This is the first time I'm building with the help of AI (Claude), but I don't really let it write code. It has been really helpful in discussing ideas (I'm mostly working alone), like a rubber duck that actually talks back. The code it generates however has many subtle and sometimes obvious mistakes in them, so I decided to write the code myself.
The main benefits over the previous version will be:
- Dynamic reports (ClickHouse queries are generated on the fly)
- Better performance
- Much cleaner and less code
- Better and faster data ingestion including new features, like full support for JSON metadata
- And much more
It feels kinda crazy that I've been working on this project for more than 5 years now and it's still exciting :)
[0] https://github.com/pirsch-analytics/pirsch (v7 is the new branch)
[1] https://pirsch.io/