From paper to runnable code in 60–120 seconds

Upload a paper.
Run the code.

RunPaper reads any ML research paper and gives you a complete Python scaffold — architecture, training loop, config — plus an interactive diagram and a reproducibility audit. No more reverse-engineering math into code.

One free paper upload · No credit card · 60 second setup

How it works

Five AI pipelines run in parallel in the background

1

Upload any ML paper PDF

arXiv papers, conference submissions, preprints — anything with text.

2

AI extracts & generates

5-step pipeline: structure extraction → code scaffold → reproducibility audit → architecture flowchart → Q&A pairs. Takes 60–120 seconds.

3

Explore, learn, and run

Interactive diagram, navigable code, PDF side-by-side, and a chat assistant — all grounded in your specific paper.

Everything in one place

Six tabs. One uploaded PDF.

Learn — Architecture diagram

Interactive ReactFlow graph of the model's data flow. Click any node to see its math, description, and exact code snippet.

Code — Runnable scaffold

model.py, train.py, config.yaml, requirements.txt — all generated from the paper. # TODO markers flag what the paper leaves ambiguous.

Paper — Inline PDF

Original PDF rendered alongside your code and diagram. No more tab-switching between your browser and your editor.

Extraction — Structured data

Title, authors, key equations (LaTeX rendered), hyperparameter table with descriptions, and clickable dataset links.

Reproducibility — Audit

20-point checklist: ✅ specified in paper vs ❌ missing — with suggested defaults for everything underdefined.

Chat — Direct or Socratic

Ask questions grounded in the paper and code. Switch to Socratic mode and the AI guides your thinking instead of just answering.

Not just another paper reader

Most tools help you understand papers. RunPaper helps you run them.

Other toolsSummarise, explain, search papers
RunPaperGenerate code you can actually execute
Other toolsFlat text answers to questions
RunPaperAnswers with code refs + flowchart nodes
Other toolsNo reproducibility signal
RunPaper20-point audit with suggested defaults
Other toolsRead the architecture in prose
RunPaperClick through the architecture interactively

Start with one paper — free

No account needed for your first upload. Sign in when you're ready for unlimited access.