GENESIS
GENESIS transforms academic research papers into interactive, runnable prototypes in under 60 seconds.
Core trick
Extracts equations and parameters from an arXiv page in one model call, then runs generated JavaScript in a Sandpack sandbox with sliders.
Description
GENESIS transforms academic research papers into interactive, runnable prototypes in under 60 seconds. The problem: researchers publish thousands of papers weekly, but almost none include working code. Understanding a paper means manually decoding equations and rewriting algorithms from scratch — a process that takes days or weeks.
GENESIS solves this with a single URL input. Users paste an arXiv link, and Claude Opus 4.7 extracts the paper's "Paper DNA" — title, classification, core equations, tunable parameters, and optimal visualization type. Then a second-stage high-effort prompt generates the actual algorithm as pure JavaScript. The code runs instantly in a sandboxed Sandpack executor with interactive parameter sliders that update the visualization in real-time.
We also built an AuditWorker using Anthropic's Managed Agents that automatically verifies the paper's numerical claims against the generated code, plus a Debate mode where two Opus 4.7 instances evaluate code fidelity.
README excerpt
<div align="center"> # GENESIS ### From Research Paper to Working Prototype in 60 Seconds. [](https://nextjs.org) [](https://www.anthropic.com) [](https://www.typescriptlang.org) [](https://tailwindcss.com) [](https://sandpack.codesandbox.io) **Paste any research paper. Watch it become an interactive simulation. Stress-test the methodology with autonomous AI reviewers.** </div> --- ## TL;DR GENESIS is an AI-powered scientific simulation platform. You give it a paper — an arXiv link, a PMC URL, an Elsevier paywall page, even a natural-language research problem — and roughly 60 seconds later you're scrubbing parameter sliders on a runnable JavaScript implementation of that paper's core algorithm. Two Opus 4.7 agents will then debate the methodology in real time. A third will run a reproducibility audit against the paper's own numerical claims. A fourth will sweep the parameter space and report the regimes where the algorith