Neura Market
GitHub logo

GitHub

Free

Backward-traceable AI-driven scientific research

FreeFree tier
Type
Open Source
Company
Technion-Kishony-lab
Visit Website GitHub (40 stars)

About GitHub

data-to-paper is an automation framework that systematically navigates interacting AI agents through a complete end-to-end scientific research process, starting from raw data alone and concluding with transparent, backward-traceable, human-verifiable scientific papers. It is field-agnostic and supports both autonomous operation (autopilot) and human-guided mode (copilot) via a graphical interface. The framework includes coding guardrails to minimize common LLM errors and creates manuscripts where numeric values can be click-traced to the code that generated them. Developed by Technion-Kishony-lab and described in a NEJM AI paper, it covers data exploration, literature search, hypothesis generation, data analysis, interpretation, and step-by-step paper writing.

Key Features

End-to-end field-agnostic research from data exploration to paper writing
Traceable 'data-chained' manuscripts with clickable numeric values
Autopilot (fully autonomous) or Copilot (human-guided) modes
Coding guardrails that override statistical packages to minimize LLM errors
Integrated literature search and hypothesis generation
Rewind and replay functionality for process control

Pros & Cons

Pros
  • Produces fully backward-traceable manuscripts for human verification
  • Supports both autonomous and human-guided workflows
  • Field-agnostic design works with any raw or annotated data
  • Coding guardrails reduce common LLM statistical errors
  • Open-source and freely available (MIT license)
Cons
  • Requires API keys and access to LLM services (costs may apply)
  • Output quality depends on the underlying LLM model
  • May require significant computational resources for large datasets
  • Autonomous mode may still need human oversight for reliability

Best For

Automated scientific research and paper generationData analysis and interpretation with traceabilityReproducible and verifiable AI-driven studiesEducation and demonstration of LLM capabilities in researchRapid prototyping of hypotheses from raw datasets

FAQ

How do I install data-to-paper?
Run 'pip install data-to-paper' and then execute 'data-to-paper' from the command line. See INSTALL.md for dependencies.
Is data-to-paper open source?
Yes, the code is available on GitHub under the MIT license (LICENSE file included).
Can I guide the research process manually?
Yes, the platform supports a Copilot App that allows users to oversee, inspect, and guide the research, set goals, provide reviews, and rewind steps.
What kind of research can it handle?
It is field-agnostic and can work with any raw data, from data exploration to paper writing, including literature search and hypothesis generation.