A time for (un)learning
Jun 02, 2024I quit my job. It was a great job, but in my defense, I have a thing for quitting great jobs. This time it hits different, though.
I’ve spent the last few years building software for well-funded startups with a similar setup: a bunch of services that may or may not talk to each other, running in a managed k8s cluster on <your-favorite-big-cloud>. I’m not going to deny the comfort and benefits of running your infra on auto-pilot, shit’s good. At the same time, I’ve seen firsthand how “throwing money at the problem” for too long can become a habit that’s difficult to unlearn. I started feeling the need to scale things down and play with new toys.
Most importantly, I’ve recently been experiencing a strong burst of creativity and just couldn’t find enough time to fully indulge in it.
LLMs are getting tiny and weirdly powerful, companies are bouncing back from the unfulfilled promises of microservices, SQLite is now the database your girlfriend tells you not to worry about (really), React has discovered PHP and the industry has come up with a whole new set of buzzwords, my favorite one is RAG, which is a fancy acronym for “query your database first so ChatGPT doesn’t make things up”. The list goes on.
There’s just too much fun stuff happening right now and I’m itching to build things just for the sake of building things.
So. What am I going to be doing with all this free time? A bit of everything:
- Releasing a mini PaaS called watergun, initially targeted at single-node workloads. I’ll be open sourcing this within the next couple of weeks.
- Skateboarding as much as my legs allow.
- Publishing on this blog from time to time (and migrating many thoughts from the past).
- Learning more about vector databases by building a product with sqlite-vec.
- Starting and growing a (slightly NSFW) Instagram meme account.
- Exploring smol LLMs — think running TinyLlama on a Raspberry Pi 5 until AGI is achieved.
That’s all I can think of for now.