NumPy Dojo

NumPy Dojo is a browser-based NumPy learning platform I spec'd and shipped with AI-assisted implementation. Most tutorials require a local Python environment; this runs a custom in-browser engine so anyone can open the site and practice immediately. I directed the build — lesson scope, UX, and architecture — rather than hand-writing every line of the engine myself.

Learning NumPy typically means setting up Python, installing packages, and configuring a local environment before writing a single line of code. That setup friction is a real barrier — especially for people who are new to Python or just want to explore. Existing browser-based options (like Google Colab) are overkill for focused NumPy practice.

• In-browser NumPy engine

  Problem

  Learning NumPy required installing Python and packages first.

  Fix

  AI-assisted build of a client-side engine targeting a curated np.* subset for lessons — intended to reject unimplemented ops clearly. I directed the architecture; I haven't audited every op boundary myself.

  Result

  Anyone could run NumPy-style syntax in the browser without a local install.

• Progressive lessons

  Problem

  Tutorials jumped around without a path from arrays to linear algebra.

  Fix

  22 progressive lessons with a built-in editor and automated output validation.

  Result

  Learners could practice and get checked without leaving the page.

• Real-world scenarios

  Problem

  Lessons showed syntax but not when or why to use NumPy.

  Fix

  12 scenarios across data analysis, finance, image processing, and engineering.

  Result

  Practice was tied to actual use, not just API trivia.

• Quiz system

  Problem

  There was no way to test retention beyond running lesson code.

  Fix

  Configurable quizzes (10–25 questions), mixed formats, retries, and history.

  Result

  Learners could measure themselves and see past attempts.

• Progress tracking

  Problem

  Refreshing the browser meant losing code and any sense of completion.

  Fix

  Completion meter, localStorage persistence, shortcuts, and adjustable editor font.

  Result

  Progress and code survived sessions without an account.

• PostHog analytics

  Problem

  No product or error signal from real learner usage.

  Fix

  PostHog on client and server for tracking and error monitoring.

  Result

  Usage and failures were visible after launch.

• CI/CD

  Problem

  Deploys and checks needed to be automatic for a solo project.

  Fix

  GitHub Actions CI/CD deployed to Vercel with zero config.

  Result

  Main stayed shippable without a manual release ritual.

Live and open-sourced for my AI bootcamp learning and other learners. No public usage metric claimed. Custom in-browser engine vs heavier options like Pyodide was the intended tradeoff (faster load, lesson-scoped control) — I directed that decision but didn't implement the engine solo.