research-and-visualize

Multi-audience interactive explainers, each built with the same pipeline: deep research → layered markdown corpus (🧒 intuition / 🎓 technical / 🔬 frontier) → build-from-scratch code thread → interactive single-page site (distill.pub / byhand.ai / makingsoftware.com / Karpathy styles) → independent factual + widget-faithfulness review.

The pipeline itself lives in skill/ — a Claude Code skill. To install it, symlink it into your skills directory:

ln -s "$(pwd)/skill" ~/.claude/skills/research-and-visualize

Then invoke with /research-and-visualize <topic> in Claude Code.

Topics

Folder	Topic	One-liner
jepa-world-models/	JEPA & world models	Predict in embedding space, not pixel space — LeCun's bet, from I-JEPA to LeJEPA
orbital-mechanics/	Orbital mechanics	An orbit is falling sideways fast enough to keep missing the ground
quantum-computing/	Quantum computing	Steer probability amplitudes so wrong answers cancel and right answers add
elasticity/	Elasticity	How materials stretch, bend, and snap
electromagnetism/	Electromagnetism	Fields, charges, and Maxwell's unification

Each explainer contains

<topic>/
├── README.md          # TOC + reading orders
├── topics/            # layered markdown (intuition / technical / frontier)
├── code/workshop/     # build-from-scratch thread: naive → fix → real data → modern variant
├── glossary/
├── papers/ or references/
└── site/              # interactive explainer — static, no build step

Viewing a site

python3 -m http.server 8765 --directory <topic>/site
open http://localhost:8765/

Running the code

cd <topic>/code
python3 -m venv .venv && source .venv/bin/activate
# dependencies are listed in each topic's code/README.md
# (most need only numpy; jepa-world-models needs torch + torchvision + matplotlib)

License

Dual-licensed: code under MIT, prose and educational content under CC BY 4.0.