Build an AI stock predictor. Then find out if it actually works.
Go from high-school math and a little Python to a working LSTM model in 8 modules. Learn the math by playing with it, let AI write code for you, and learn to test it like a professional. Everything runs in your browser.
Every module mixes four ways to learn
Lecture
Short lessons and slides. No jargon without a picture.
Discover
Playful widgets: drag a neuron's weights, spot a data leak, beat a coin flip.
Quiz
8 to 10 questions with explanations. Retake as often as you like.
Lab
Real Python notebooks in the page. Each lab adds a part to your predictor.
8 modules, one predictor
Each lab builds a piece you reuse in the final project.
- 01 · 4 h
Python for AI
NumPy, pandas and plots. Load real prices and chart a moving average.
Discover: Array Arcade - 02 · 6 h
Neural network math
From the slope of a line to backpropagation. Build a network from scratch.
Discover: Neuron Playground - 03 · 3 h
Setting the goals
What to predict, how to score it, and the baselines you must beat.
Discover: Beat the Coin - 04 · 5 h
Preparing data
Clean, scale and window time series without peeking at the future.
Discover: Spot the Leak - 05 · 4 h
Writing code with AI prompts
The spec-first prompt. Get AI to write code you can understand.
Discover: Prompt Duel - 06 · 4 h
Verifying the code
Tests, sanity checks and leak detectors for AI-written code.
Discover: Bug Hunt - 07 · 6 h
Building an LSTM
Memory with valves. Build one LSTM cell, then train a full model.
Discover: Memory Valves - 08 · 5 h
Capstone
Your own forecaster, tested fairly and published with a model card.
Discover: Forecast Week
My Stock Forecaster
Pick a stock, build an LSTM that predicts its next-day move, and prove with a fair test whether it beats simple guesses. Then lock in five real predictions during Forecast Week.
You pass by testing honestly, not by getting rich. A model that loses to the baseline but tells the truth passes. One that "wins" by peeking at the future doesn't.
What you hand in
- A notebook that runs top to bottom, every time
- A goal card: target, horizon, metric, baselines
- Results on 12 months your model never saw
- A one-page model card with an honest verdict
- A prompt log of AI code you checked and fixed
- Five Forecast Week predictions
Pass mark 70/100 · Bonus badge for beating the baseline