Paper → Code Engine

Paste a paper.
Get working code.

Turn any arxiv research paper into a runnable Python implementation. Stop re-reading methodology sections — start running experiments.

Free to start 3 free generations/month Runnable Python output
generated_model.py — from arxiv:2301.08243
# Auto-generated from: "Scaling Data-Constrained Language Models"
import torch
import torch.nn as nn

class ScalingLM(nn.Module):
    def __init__(self, d_model=512, n_layers=6):
        super().__init__()
        self.encoder = nn.TransformerEncoder(
            nn.TransformerEncoderLayer(d_model, nhead=8),
            num_layers=n_layers
        )
        self.output = nn.Linear(d_model, 50257)
Three steps. Zero boilerplate.
From paper URL to running code in under a minute. No manual extraction, no guesswork.
01

Paste the arxiv URL

Drop any arxiv paper link. Shortlyst downloads the PDF, parses the full text, and identifies the core methodology and architecture.

supports arxiv + PDF upload
02

AI extracts the method

Our model reads the paper like a researcher would — identifying the architecture, loss functions, training procedure, and key hyperparameters from methodology sections.

figures + equations included
03

Get runnable Python

Receive clean, well-documented Python code with proper imports, model definitions, training loops, and inline comments referencing original paper sections.

pip install → python run.py
Built for people who read papers, not just cite them.
Whether you're reproducing SOTA or evaluating a new approach for your team — Shortlyst cuts days of implementation down to minutes.

ML Engineers

Quickly prototype architectures from new papers. Skip the boilerplate and get straight to experiments. Compare approaches side-by-side with generated code.

🔬

Researchers

Reproduce paper results without spending weeks on implementation. Validate methodology claims with working code. Build on prior work faster.

🏢

AI Teams

Evaluate SOTA approaches before committing engineering resources. Get working prototypes for architecture reviews. Accelerate your research-to-production pipeline.

Manual implementation is a time sink.
The average ML paper takes 2-5 days to implement from scratch. Most of that time is parsing methodology, not writing code.
The old way

Read → Implement → Debug

  • 2-5 days per paper implementation
  • Hours parsing dense methodology sections
  • Missing details that require reading appendices
  • Bugs from misinterpreting equations
  • No standardized code structure
With Shortlyst

Paste → Generate → Run

  • Working code in under 60 seconds
  • AI extracts methodology automatically
  • Handles equations, figures, and references
  • Clean, documented, runnable output
  • Consistent PyTorch code patterns
Start free. Scale when you need to.
No credit card required. Upgrade only when you need more papers per month.
Free
For trying it out
$0 /mo
Free forever
  • 3 papers per month
  • Python code output
  • arxiv URL support
  • Basic model architectures
Get started free
Team
For AI research teams
$99 /mo
Billed monthly
  • Everything in Pro
  • 5 team members
  • Shared paper library
  • Custom code templates
  • API access
Start Team →
Common questions
Everything you need to know before pasting your first paper.

What types of papers work best?

ML/AI papers with clear methodology sections work best — think model architectures, training procedures, and novel algorithms. Papers with strong mathematical notation and clear pseudocode produce the best output.

Is the generated code production-ready?

It's prototype-ready. The code is clean, documented, and runnable — but it's meant for experimentation and validation, not direct production deployment. Think of it as a massive head start.

Which frameworks does it support?

Output is primarily PyTorch, since that's what most research uses. We generate standard Python with pip-installable dependencies. JAX and TensorFlow support are on the roadmap.

Can I upload PDFs directly?

Yes, on Pro and Team plans. Free users can paste arxiv URLs, and we'll fetch the paper automatically. Pro users can upload any research PDF directly.

How accurate is the extraction?

For well-structured ML papers, we capture the core architecture and training procedure with high fidelity. We include inline comments linking back to paper sections so you can verify every decision.

Is it really free to start?

Yes. 3 papers per month, no credit card required. We want you to see the quality before you commit. The free tier never expires.

Stop reading papers.
Start running them.

Paste your first arxiv link and get working code in under a minute. No signup. No credit card. Just results.

Try it now — it's free →