Welcome to the LuzTech Blog

Welcome to LuzTech’s official tech blog!

I built this space to share a bit of the practical knowledge that LuzTech’s engineering team has picked up along the way.

The idea is to share ideas, curiosities, and lessons learned from the team’s day-to-day work, including projects, architectural evolutions, and engineering decisions.

I also plan to share interesting curiosities and projects I come across in the community.

And for the very first post, I’d like to walk you through how I built this blog.

Table of contents

The requirements

My first requirement, which set the direction of the entire blog, was that it had to be static — that is, no backend.

Wait, a backend developer building a project without a backend?

Yep. One important lesson I’ve learned throughout my career is to pick the right tool for the right job.

Thinking about my current scenario — being the company’s sole developer and the blog’s only author — having a backend would only bring me more work and cost.

And that wasn’t the goal! The goal was a simple, cheap blog, easy to write content for (pnpm post:new) and easy to publish (CI/CD on GitHub). A backend would force me into a trade-off between cost and convenience:

1. Azure Webapp: Good and reasonably fast, but with a cost higher than I’m willing to pay right now

2. Azure Functions: Very low cost (as long as the blog didn’t blow up, which I don’t see happening initially), but with poor performance, since cold starts are inevitable (and brutal for this kind of application)

So I figured it was better to go with SSG, or Static Site Generation, which would give me a simple, fast, and easy-to-maintain blog. The downside here is if I ever want more complex features, like comments, many authors, RBAC, etc. I will need to migrate. But since that’s not the case (at least now), SSG works just fine for me.

After chatting with my imaginary friends (ChatGPT and Gemini), I concluded that Astro would be the best tool for the job.

But why Astro and not <insert your favorite framework here>?

Well, because on top of SSG, Astro lets me use MDX, which allows me to do things like drop React’s famous counter on the very next line:

And the integration between text and components is really good, so it ended up winning.

Theme

After that came my search for a theme I actually liked. I went to the Astro themes page for blogs and, to be honest, the current theme wasn’t the first one to catch my eye — that was Vector.

But, since nothing in life is perfect, Vector is a paid theme. And since the blog’s initial requirement was to be cheap, I went with Astro Paper, which is free and fits my needs well.

Astro Paper is no slouch either. It has a more minimalist vibe, which I think fits the blog’s proposal nicely:

With the theme picked, it was time to ✨customize✨.

Customization

A key thing that helped me a lot at this stage was the existence of the LuzTech Typography repository. It allowed me to keep a visual identity consistent with the rest of the company, without having to reinvent the wheel.

And, of course, I wasn’t the one who did most of the customization — that was GitHub Copilot.

What do you mean? You used Artificial Intelligence to write code? Didn’t it end up full of slop?

Good question, my friend, and the answer is: yes, I used GitHub Copilot to generate a good chunk of this blog’s frontend, including the components used by the post (not the text though — that I still write the old-fashioned way).

Now, about the slop… Yes, it produced some slop, but surprisingly, not that much. See, I used one of the most powerful models available right now: Claude Opus 4.7, which generally writes code pretty reasonably. And, of course, I was testing and reviewing throughout the process, and asking for things little by little.

Artificial Intelligence doesn’t replace a person, it replaces a tool — or better yet, it is a tool, just another one, like compilers, linters, IDEs, etc. And, like any tool, it needs to be used with care and attention.

That said, I referenced the typography project in my prompts and was very clear about what I wanted to be implemented. Copilot often delivered something that wasn’t exactly what I wanted, but with a few tweaks and clearer prompts, it kept delivering what I needed, bit by bit.

Conclusion

It was a fun experiment, to be honest. It’s not the most over-engineered blog in the world, but sometimes, less is more than enough.

Hope you enjoyed it!