I tried to fine-tune AI models… it was painful. So I built this
I’ve been building a small project called Miniclay AI.
The idea came from a frustration I kept hitting over and over.
Every time I tried to fine-tune a model, I had to:
- find datasets
- clean messy data
- write training scripts
- debug notebooks
- deal with GPUs
Even with tools like Hugging Face or Colab, it still felt messy and slow.
The biggest realization:
👉 training isn’t the hard part — data is.
Cleaning data, fixing structure, removing noise… that’s where most time goes.
So I started building something for myself:
A system where I could just describe the model I want.
Example:
“AI that generates product descriptions from images”
And it would:
- find relevant datasets
- clean them automatically
- run the training pipeline
- give me a fully fine-tuned, ready-to-use model
That’s how Miniclay AI started.
---
Right now:
- Free users → get a notebook (DIY mode)
- Paid users → get cloud fine-tuning (no setup)
The idea is:
Let beginners learn the hard way, and let builders skip it.
---
I’ve shown it to a few ML devs and the reaction was:
“This is exactly the pain point.”
Which was reassuring… but also scary 😅
---
I’m launching it publicly in a few days.
Before that, I’d love honest feedback:
👉 Would you use something like this?
👉 What would stop you from trusting it?
👉 Is this actually solving a real problem, or am I overthinking it?
---
Would really appreciate blunt feedback 🙏
