How karpathy/scholaroctopus works
File structure
• Top level:
• README.md – project description, visualization goals, contribution hints
• render/ – contains all client artifacts
– data5.json – the precomputed 2D embedding and metadata for papers
– index.html – the main visualization page (HTML, CSS, JS all together)
• No backend folder, no build scripts, no tests; it’s purely a static site.
README for additional insights
• Primary visualization uses t-SNE and tf-idf to embed paper text.
• The “render/index.html” is the canonical view; contributions add new HTML views.
• No database or dynamic API—the data is baked into a JSON file.