NeuralPath

NeuralPath is a local, self-paced AI/ML bootcamp delivered as a Windows desktop app. It walks you from mathematical foundations to research-level topics over a 21-week curriculum, pairing daily theory lessons with adaptive quizzes, one medium-sized project per week, and full progress tracking — streaks, mastery scores, weaknesses, and time spent. The end goal is a Portfolio Export you can present to companies and researchers.

The app runs entirely offline for lessons and projects. It uses the Claude API only for three things: generating quiz questions, grading short answers, and reviewing your project submissions. Your API key is stored locally and never leaves your machine except in calls to Anthropic.

---

Prerequisites

• Node.js 20 LTS (recommended) — see the toolchain note below before using newer majors
• Python 3.11+ (for running ML project code locally)
• An Anthropic API key — get one at console.anthropic.com

Windows native-build toolchain (required)

NeuralPath uses better-sqlite3, which is a native C++ addon. On npm install it loads a prebuilt binary if one exists for your Node version, otherwise it compiles from source — and compiling needs a C++ toolchain. Two ways to satisfy this:

1. Use Node 20 LTS. It has prebuilt better-sqlite3 binaries, so npm install works with no compiler.
2. Or install the C++ build tools (needed for very new Node majors like 22/24, which may lack prebuilts): install Visual Studio Build Tools with the "Desktop development with C++" workload, then re-run npm install.

Symptom of a missing toolchain: npm install fails with gyp ERR! find VS ... could not find a version of Visual Studio. Because this is an Electron app, the addon is also automatically rebuilt against Electron's ABI during npm run build once a compiler is present — no manual electron-rebuild step needed.

Setup

npm install
pip install -r python/requirements.txt
python assets/datasets/prepare_datasets.py   # generates iris.csv, wine.csv, ecommerce.db
npm run dev

Add your API key during the in-app onboarding wizard (or copy .env.example to .env and fill in ANTHROPIC_API_KEY). It is saved to your user data directory, never hardcoded.

Build a Windows installer (.exe)

node scripts/genIcon.mjs   # regenerate assets/icon.png if missing
npm run build

The NSIS installer is written to dist/.

Useful scripts

---

Curriculum Overview

Heavy weeks that need a GPU (CNNs, object detection, segmentation, LoRA, GANs, RL) integrate with Google Colab: NeuralPath generates a ready-to-run notebook, you train on a free T4 GPU, then import a results.json back into the app so progress tracking stays intact. See the in-app Colab Guide.

Research alignment

Phases 4–5 are framed around the communication-aware ML research program of Prof. Deniz Gündüz and Emre Ozfatura (Imperial College London). Relevant weeks surface a "Research Extension" panel and optional research-track challenges; your Portfolio collects the papers you engaged with as PhD-trajectory work. Source papers are in papers/.

---

Architecture

• Electron shell + React 18 / Vite renderer, styled with Tailwind and a custom dark design system.
• better-sqlite3 for all local state (topics, sessions, quizzes, projects, streaks, goals).
• Anthropic SDK runs only in the Electron main process; the API key is never exposed to the renderer.
• A bundled Python runner (python/runner.py) executes project code via child_process.

electron/   main process, preload bridge, SQLite, Claude, notebook generator
src/        React app — pages, components, Zustand store, curriculum
python/     project runner + requirements
assets/     app icon + datasets
scripts/    db reset + icon generation

Screenshots

Dashboard, Lesson, Quiz, Project, Progress, and Portfolio screenshots go here.

License

MIT