ResearAI/AutoFigure — reverse-engineered prompt

Build me a web app that turns scientific text or a research paper PDF into a clean publication ready diagram.

I want users to paste a description, upload a paper, choose a topic like paper, survey, blog, or textbook, then generate a figure. The app should use an LLM to create the first diagram, review it, give itself a score, and refine it until the result is good enough. Show the progress so the user can see each improvement round.

The final output should be downloadable as SVG and also as draw.io compatible XML. If possible, include an optional image enhancement step that can make a few prettier versions of the figure. Add a simple settings page where I can enter API keys, provider, model, number of iterations, and output format.

Please make it easy to run locally with a frontend, backend, and clear start and stop commands. Look up current docs online if you need to.