Python to AI — Interactive Learning Platform

A free, self-hostable tutorial platform taking you from Python basics to generative and adversarial AI — with live in-browser ML model demos powered by ONNX Runtime Web. No server required; everything runs in your browser.

Live: ikathuria.github.io/python-to-ai

---

Learning Path

#	Lesson	Topics	Try it in the browser
—	History of AI	80 years of AI, hype vs. reality (no code)	Animated timeline
00	Python basics	Data types, structures, functions, OOP	—
01	ML basics	NumPy, Pandas, data preprocessing	—
02	Supervised learning	Classification, regression, KNN, decision trees	Live predictions (ONNX)
03	Unsupervised learning	K-means clustering, PCA	Cluster assignment (ONNX)
04	Recommendation systems	Content-based & collaborative filtering	Movie recommender (ONNX)
05	Deep learning	PyTorch, backprop, neural networks	Gradient descent & activation playgrounds
06	Computer vision	CNNs, convolution, image classification	Colour classifier (ONNX)
07	Time series	Stationarity, ARIMA, LSTM forecasting	Moving-average smoother
08	Language (NLP)	Word2Vec, PMI, language models	Cosine similarity explorer
09	Generative AI	Next-token prediction, Baby GPT, RAG, diffusion	Be the language model
10	Adversarial AI	Adversarial examples, FGSM, prompt injection, defences	Fool a classifier

---

Repo Map

Path	What's in it
index.html, app/pages/	The website (static HTML, served by GitHub Pages)
app/onnx_models/	Trained models the site runs in the browser
app/styles/	Brand tokens, shared styles and scripts, built lesson CSS, episodes.js (video list)
0 Basic_Python_Concepts/ … 8 Adversarial Threats/	Notebooks behind each topic. Each folder has a README linking to its lesson
scripts/	Export models to ONNX, keep the nav in sync, make share images
tests/	Site checks that run in CI
CLAUDE.md	Conventions for working on the repo (datasets, adding lessons)
PLAN.md	Roadmap
VIDEOS.md	YouTube episode plan linked to each lesson
BRAND.md	Brand guidelines
design/	Brand exploration: the 8 directions and the brand sheet

Datasets aren't stored in the repo: notebooks download them from Kaggle on first run with kagglehub (public datasets, no account needed).

---

Features

• Live demos: 15 ONNX models plus small JavaScript demos run entirely in the browser (no server, no Python needed)
• Learning path: lessons grouped into four stages, with a progress bar and "Continue" button saved in your browser
• Knowledge checks: a quiz at the end of every lesson; a perfect score marks the lesson complete
• Copy buttons on every code block and "Try it yourself" exercises
• Light and dark mode, mobile friendly

---

Tech Stack

Layer	Choice
Frontend	Static HTML/CSS/JS; lesson styles compiled with Tailwind CSS
Model inference	onnxruntime-web (runs in the browser)
Syntax highlighting	Highlight.js
Hosting	GitHub Pages

---

Running Locally

View the site (no build step, no Node.js):

git clone https://github.com/ikathuria/python-to-ai.git
cd python-to-ai
python -m http.server 8000
# open http://localhost:8000

Run the notebooks (macOS, Linux or Windows):

pip install -r requirements.txt
jupyter notebook

Working on the site:

python scripts/sync_nav.py   # after adding/reordering lessons in scripts/lessons.py
npm install && npm run build:css   # after changing Tailwind classes in app/pages/*.html
pip install -r requirements-ci.txt && pytest   # the same checks CI runs

---

Re-training / Exporting Models

The ONNX models in app/onnx_models/ were exported from the notebooks. To re-export them:

python scripts/export_tutorial_onnx.py     # scikit-learn models
python scripts/export_recommender_onnx.py  # movie recommender
python scripts/export_colour_cnn.py        # colour CNN (needs torch)

---

For more tutorials

• Videos on YouTube
• My website: ishani.kathuria.net
• Connect on LinkedIn
• Articles on Medium
• GitHub Wiki