projects / paper-plot
paper-plot.
AI-native best practice for academic figures. The figure you wanted, with every piece editable.
The problem.
Researchers spend hours in PowerPoint, draw.io, matplotlib, or TikZ to make figures that mostly need to look right — and inevitably need one more edit the night before the deadline.
Generative models can sketch a figure in seconds, but the output is a flat raster. Once you want to change a single label, you're back to redrawing from scratch.
The recipe.
paper-plot is a small suite of tools built on a single recipe:
- L1 · Generate: a structured prompt yields a strong first-draft raster.
- L2 · Structure: the raster is segmented into addressable layers (boxes, arrows, labels).
- L3 · Render: each layer becomes editable SVG you can tweak directly.
What's in it.
- sketch — flowcharts, architecture diagrams, teaser figures. Live demo, alpha quality.
- chart — line plots, bar charts, ablation curves. Design page only for now.
Why it matters.
The figure is where the paper meets the reader. Most figures are made under time pressure with the wrong tool for the job. paper-plot is a small bet that the right tool is one that gives you both generative polish and structural control — neither alone is enough.