I’m 14 and I Vibe-Coded an AutoML Platform Because sklearn Made Me Rage Quit

TL;DR: I was tired of writing 50 lines of ML boilerplate. So I built Velo AI — type what you want to predict, upload a CSV, get a trained model + API in ~20 seconds. Free demo link at the bottom.

The problem: I’m 14, impatient, and lazy

I started Python at 9 because my dad forced me into a course. I forgot OOP completely and came back at 14 this year.

First thing I tried? Machine learning. Every single tutorial looked like this:

python
import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split, GridSearchCV
from sklearn.preprocessing import StandardScaler, OneHotEncoder
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
from sklearn.ensemble import RandomForestRegressor
from sklearn.metrics import mean_squared_error, r2_score
import joblib

#Load data
df = pd.read_csv('data.csv')

#... 40 more lines of preprocessing hell
#... before you even train one model

I just wanted to predict student exam scores. Not spend 3 hours googling ValueError: could not convert string to float.

So I rage quit sklearn tutorials and vibe-coded my own solution.

What I built: Velo AI
Velo = Prompt → Trained ML Model

You literally type in plain English:

Code
predict house prices from sqft, bedrooms, location
Then upload your CSV. Velo handles the rest:

LLM figures out the task: Regression vs classification vs clustering
Auto data cleaning: Missing values, categorical encoding, outlier detection
Auto model selection: Trains LogisticRegression, RandomForest, XGBoost, etc. Runs cross-validation and picks the winner
Exports everything: ONNX model file + FastAPI inference code + accuracy report + feature importance graph
Total time: ∼20 seconds.

Total lines of code YOU write: 1 sentence.

Demo
[Image blocked: Velo Demo GIF]
^ Me going from idea → trained model faster than you can say pipeline.fit()

Try it free right now: [Your Hugging Face Space Link Here]

No sign up. No credit card. No email. I'm 14, I don't even have Stripe or a bank account to charge you lol.

Tech stack for the nerds
I built this in 2 weeks during summer break using:

Component

Tech

Why

Backend

FastAPI + Python 3.11

Fast, async, easy

LLM Brain

Groq with Llama-3.1-70B

Parses your prompt into ML task

AutoML Engine

scikit-learn + AutoGluon

Model training + HPO

Model Export

ONNX Runtime

Run predictions anywhere, no Python needed

Frontend

Gradio on HF Spaces

Because it’s free and I’m broke

Cost

$0/month

Student budget = $0

The whole thing runs on Hugging Face’s free tier. If this blows up I’ll have to figure out money stuff, but for now it’s 100% free.

Why I’m posting this on Dev.to
This is v1. It definitely has bugs. It probably breaks on weird datasets. The code is probably messy.

I need you to roast me.

I’m 14 so please be nice, but also be real. I want to make this actually useful for students, indie hackers, and anyone who hates boilerplate as much as I do.

Tell me:

What’s the dumbest part of my architecture?
What feature is missing that would make you actually use this?
Would you trust a model trained by a 14yo? Be honest.
Drop a comment. I’ll reply to every single one.

What’s next if people actually use this
If 100 people try Velo and don’t hate it, I’m building next:

Auto-EDA Reports: Upload CSV → get a PDF with graphs, correlations, data warnings
One-click API Deploy: Trained model → public endpoint in 1 click
Time Series Support: "forecast sales for next 30 days" prompts
Model Explanations: SHAP values explained like you’re 5
"Fix my CSV" button: LLM cleans your messy data automatically
But only if this isn’t a complete dumpster fire. You tell me.

Link again for people who scrolled fast
Free demo, no login: I just wanted to predict student exam scores. Not spend 3 hours googling ValueError: could not convert string to float.

So I rage quit sklearn tutorials and vibe-coded my own solution.

What I built: Velo AI
Velo = Prompt → Trained ML Model

You literally type in plain English:

Code
predict house prices from sqft, bedrooms, location
Then upload your CSV (or if you don't have one, then Velo auto generates datasets for you!!!). Velo handles the rest:

LLM figures out the task: Regression vs classification vs clustering
Auto data cleaning: Missing values, categorical encoding, outlier detection
Auto model selection: Trains LogisticRegression, RandomForest, XGBoost, etc. Runs cross-validation and picks the winner
Exports everything: ONNX model file + FastAPI inference code + accuracy report + feature importance graph
Total time: ∼20 seconds.

Total lines of code YOU write: 1 sentence.

Try it free right now: https://huggingface.co/spaces/its-leealis/velo-ai

No sign up. No credit card. No email. I'm 14, I don't even have Stripe or a bank account to charge you lol.

Tech stack for the nerds
I built this in 2 weeks during summer break using:

Component

Tech

Why

Backend

FastAPI + Python 3.11

Fast, async, easy

LLM Brain

Groq with Llama-3.1-70B

Parses your prompt into ML task

AutoML Engine

scikit-learn + AutoGluon

Model training + HPO

Model Export

ONNX Runtime

Run predictions anywhere, no Python needed

Frontend

Gradio on HF Spaces

Because it’s free and I’m broke

Cost

$0/month

Student budget = $0

The whole thing runs on Hugging Face’s free tier. If this blows up I’ll have to figure out money stuff, but for now it’s 100% free.

Why I’m posting this on Dev.to
This is v1. It definitely has bugs. It probably breaks on weird datasets. The code is probably messy.

I need you to roast me.

I’m 14 so please be nice, but also be real. I want to make this actually useful for students, indie hackers, and anyone who hates boilerplate as much as I do.

Tell me:

What’s the dumbest part of my architecture?
What feature is missing that would make you actually use this?
Would you trust a model trained by a 14yo? Be honest.
Drop a comment. I’ll reply to every single one.

What’s next if people actually use this
If 100 people try Velo and don’t hate it, I’m building next:

Auto-EDA Reports: Upload CSV → get a PDF with graphs, correlations, data warnings
One-click API Deploy: Trained model → public endpoint in 1 click
Time Series Support: "forecast sales for next 30 days" prompts
Model Explanations: SHAP values explained like you’re 5
"Fix my CSV" button: LLM cleans your messy data automatically
But only if this isn’t a complete dumpster fire. You tell me.

Link again for people who scrolled fast
Free demo, no login: https://huggingface.co/spaces/its-leealis/velo-ai

If it saved you time, leave a comment. If it crashed, leave a comment. If my Python looks like it was written by a 14yo… well, it was. Leave a comment.

Thanks for reading this far. This community is the only reason I’m not banned on Reddit anymore 🙏

Are you a student who struggles with ML setup? What’s the most annoying part? Let’s complain together in the comments.

If it saved you time, leave a comment. If it crashed, leave a comment. If my Python looks like it was written by a 14yo… well, it was. Leave a comment.

Thanks for reading this far. This community is the only reason I’m not banned on Reddit anymore 🙏

Are you a student who struggles with ML setup? What’s the most annoying part? Let’s complain together in the comments.