I'm in the exact same situation as the author.

It's not productive but hell it is rewarding to dive deep into lower levels.

No regrets so far.

What was you approach? Where one starts such a project?

Early boot and early page table setup, logging single characters to simulated serial console.

Then keep adding tests/features/tests/features/...

I missed last train due to delays and there was a group of in the same situation. One nice person offered me to that I can sleep on their couch. And they were so nice to give me a ride to the station the next day.

I was so angry at first when I found out that this was my last train and I missed it but it turned out to be great story I can tell :)

Thank you strangers, I'll repay it back to somebody in the future

I'm betting against wasm and going with containers instead.

I have warm pool of lightweight containers that can be reused between runs. And that's the crucial detail that makes or breaks it. The good news is that you can lock it down with seccomp while still allowing normal execution. This will give you 10-30ms starts with pre-compiled python packages inside container. Cold start is as fast as spinning new container 200-ish ms. If you run this setup close to your data, you can get fast access to your files which is huge for data related tasks.

But this is not suitable for type of deployment Cloudflare is doing. The question is whether you even want that global availability because you will trade it for performance. At the end of the day, they are trying to reuse their isolates infra which is very smart and opens doors to other wasm-based deployments.

it's back on

but wow, it must be stressful to deal with this

The ongoing issue is the maintenance.

This can't be solved without fully trusting the LLM period.

Just don't autopilot on important code you want to own. That's good start.

Related submission to the mentioned No AI December https://news.ycombinator.com/item?id=46098433

One of the guys running the challenge here.

No cloud based AI is the hardcore version of it for sure.

I hope local models will fill 80% of use-cases so we are not tight to the big guys.

How long are you using your setup?

Author here.

I would add that doing things on your own, which is probably never optimal from time perspective, may open doors to solutions that you haven't seen before.

I recently experienced this is by not going through the beaten path - installing well-known dependencies and building solution based on that. I chose to experiment a bit and it turned out that I usually use small % of imported code.

It really changed how I look at dependencies.

I agree. If everyone always just used the standard solution we'd never get better ones. Nothing is ever done.

You can go very far with just node alone (accepts typescript without tsc, has testing framework,...). Include pg library that has no dependencies. Build a thin layer above node and you can have pretty stable setup. I got burnt so many times that I think it is simply impossible to build something that won't break within 3 months if you start including batteries.

When it comes to frontend, well I don't have answers yet.

You can write simple front-end without reactive components. Most pages are not full blown apps and they were fine for a very long time with jQuery, whose features have been largely absorbed into plain js/dom/CSS.