When you're building a side project in 2025, what's your default stack these days?
Are you still loving the reliability of Laravel or Ruby on Rails, or have you fully embraced Next.js, Bun, or something even more bleeding edge? Maybe you're mixing in tools like Supabase, Neon, or HTMX?
Curious to hear:
- What's your go-to stack for quick MVPs or weekend builds?
- Do you keep it simple or try to mirror production setups?
- What are you hosting it on?
Been thinking about this a lot while working on something for DevOps Daily and it made me wonder what others are using this year.
Drop your stack below, someone might discover their next favorite combo from your setup!
Top comments (21)
Honestly, my stack is just Math, NumPy, and built-in libs in C++/C or Python. No frameworks, no GPU. I’ve built Transformers, GPTs, Text-DDPMs (with Transformer as denoiser), and Diffusion models for vision, all running on an old AMD Phenom PC with 2 GB Memory and using under 200MB RAM. Trained the Text-DDPM in 7 minutes on 6 full-length .txt files. Way more fun than chasing big stacks.
Damn bro, that's so cool!
Appreciate it! I just enjoy pushing the limits of what's possible with minimal resources. You learn a lot more when there's nothing to rely on but fundamentals and it’s way more rewarding than spinning up another boilerplate stack. I am currently working on Diffusion model using Numpy and Math libraries and training it on 2k (2042×2042 resolution) images in seconds. I will release it soon and you will be able to run your own Image Generation AI on a toaster.
That's nice, but as a beginner myself, how do you think I can start learning the fundamentals and start implementing it, cause I just keep getting caught in "studying the theory" loop
I get you. I used to feel stuck in theory too. I’m 13, and what helped me was reading research papers and understanding the architecture first. Then I’d use one LLM to do deep research (testing several ones and choosing one), breaking down the concepts, math, training flow, everything and once I had that structured, I’d feed it into another LLM like Claude or DeepSeek to actually generate the code using only NumPy and math.
I don’t write every line myself, but I always make sure I understand what’s being built. If something breaks, I just send the traceback back to the model and let it correct it. That loop helped me build full working models like Transformers, GPTs, even Diffusion, all on a low-spec PC (because of low level maths and no bloated frameworks).
You don’t need to master everything before starting. Use the tools to help you learn while building, and feed them good inputs so you’re not guessing. That’s what made everything click for me.
I see, Thanks man!
I'll try this out!
That's seriously impressive!
Thanks! You should definitely try it too. It’s wild how much you can build with just fundamentals and the right mindset. Would love to hear if you try something similar!
Rimmel.js + CSS modules for anything frontend, especially if reactive/interactive. Vanilla JS only for vibe-coded minimalistic PoC's that are never going to grow beyond 1 hour work and nobody cares. Still looking for new serverless inspo for the back-end and for a graph db. Pulumi has got my attention, too. Enough with those declarative non-Turing-complete IaC languages...
Love the Rimmel.js pick, haven’t seen it mentioned much but it’s got potential. Pulumi’s a solid move if you want real language power in IaC. Curious what you land on for backend + graph db!
Created an example chat app with Deno Deploy, recently.
Been a great start, as it's all serverless and comes wtih its own native key-value store
For full stack stuff? Definitely Django. It's not as flexible for larger scale full stack development (unless you only use it as an API), but you can get a project started extremely quickly.
I'd probably try and use Rust for anything else though.
rust-analyzeris an absolute godsend, and I just really like the language.Django really is unbeatable for getting started fast. And yeah, Rust hits that sweet spot of performance and safety, hard to go back once it clicks. I've been using Rust for a couple of projects, the learning curve is real 😅
For little websites, you mean? Sveltekit. On Vercel or Netlify or whatever. With vanilla CSS and JS.
It's the right balance for me, if I want something reactive. It's developer-friendly.
SvelteKit is such a clean choice for small, reactive projects.
For me lately it's been Laravel Volt + Folio, Tailwind, and Postgres. Simple, fast, and easy to manage solo. Sometimes I throw in Livewire if I need reactivity without going full SPA.
Awesome, Laravel is so good!
So far, I host my projects on github pages until I learn about what else is available. Tech stacks for now are mainly React And Tailwindcss. I’m also looking into learning Next.js.
Great starting stack!
For my webapps, I'll go with angular for front, with daisyUI as design system and my backend will be powered by Quarkus/Kotlin
Postgres if I need any DB
Nice combo! Love seeing Quarkus + Kotlin. Do you usually host Postgres on Neon, DigitalOcean, or somewhere else?