I Built an Interactive Machine Learning Platform to Learn ML Visually

#python #datascience #machinelearning #opensource

When I first started learning machine learning, I noticed a pattern.

I'd read a chapter about an algorithm, watch a YouTube video, open a Jupyter notebook, read the scikit-learn documentation, search Stack Overflow when something didn't work, and then try to piece everything together myself.

Even after all of that, I still found myself asking questions like:

Why did changing one hyperparameter completely change the decision boundary?
Why does one algorithm perform better than another on the same dataset?
What actually happens while a model is training?
Why did the model make this prediction?

There were plenty of great resources, but they were scattered.

Some websites had nice visualizations but only supported a handful of algorithms. Others focused on the mathematics but weren't interactive. Most demos were static, and very few let me experiment with real datasets while seeing the impact of every change.

So I decided to build the tool I wished I'd had while learning.

Introducing Confluence

Confluence is an open-source platform for exploring machine learning through experimentation.

Instead of only reading about an algorithm, you can interact with it.

Choose a dataset, adjust hyperparameters, compare algorithms side by side, inspect predictions, and watch how models learn in real time.

Everything runs on a FastAPI backend using real scikit-learn models rather than simplified browser-only implementations.

What's included?

Currently the project includes:

38 machine learning algorithms
25 datasets (real-world and synthetic)
Interactive decision boundary visualizations
Training animations
Prediction explanations
Side-by-side algorithm comparison
Interactive metrics and evaluation tools
Algorithm encyclopedia
Python code generation based on your experiment
AI assistant for learning concepts

The goal isn't to replace notebooks or production ML libraries.

It's to make the learning process more visual and intuitive.

Check out the demo:

Built with

Next.js
TypeScript
FastAPI
Python
scikit-learn
Redis
Docker

What I learned

Building Confluence taught me much more than machine learning.

I had to think about application architecture, frontend performance, backend API design, visualization pipelines, state management, caching, deployment, and creating an experience that felt responsive even while running real ML computations.

It reminded me that building projects is one of the fastest ways to learn.