Turn the page into an experiment
I am a research assistant, and my PI assigns me papers I wish I had more intuition for. The ideas in them move, interact, and change, but the reading experience is static. Moiré turns public papers and educational articles into something readers can explore.
Prepend Moiré's domain to a source URL—no install, upload, or account required. The paper remains the primary reading surface while Moiré marks concepts worth visualizing. Readers can open prepared experiments immediately, adjust their parameters, select a focused passage to request another visualization, and pin useful results to a per-page notebook.
What it does
- Keeps experiments attached to the passages they explain.
- Speculatively pre-generates high-value visualizations so the first click feels instant.
- Supports grounded selection-to-visualization with checks for insufficient context or multiple concepts.
- Produces interactive 2D plots, animations, simulations, and constrained Three.js scenes.
- Runs generated artifacts in a validated, network-isolated iframe sandbox.
- Preserves useful experiments in a per-page notebook with links back to the source.
- Works with arXiv papers and other readable public pages, including Wikipedia.
How it works
Moiré safely fetches and sanitizes a public page, preserving arXiv's available LaTeXML structure when possible. GPT-5.6 Luna scans the source and produces ranked visualization briefs anchored to exact document selectors. Grok 4.5 generates self-contained interactive artifacts through OpenRouter, GPT-5.6 Terra provides a safe generation fallback, and GPT-5.6 Sol handles bounded structural or browser-runtime repairs. Every artifact is checked for grounding, structure, responsive layout, controls, size, and prohibited network behavior before it runs inline.
Eight preserved, prevalidated demo artifacts provide immediate examples across attention, quench dynamics, chaotic motion, and 3D Gaussian anisotropy.
Built with Codex and GPT-5.6
Moiré was developed through one primary Codex task that owned architecture, implementation, integration, review, verification, and deployment. Codex translated the Build Week epic into a dependency-aware implementation sequence, then used tests and browser QA to refine source ingestion, server-owned caching, selection guardrails, artifact sandboxing, responsive visualization contracts, and bounded repair behavior. Key product decisions—keeping the paper central and grounding every experiment in source terminology—were developed through that collaboration.
Built With
- gpt-5.6
- next.js
- openai-codex
- openrouter
- react
- three.js
- typescript
- vercel