Repolode
Submit
Gen-Verse

Gen-Verse/DiscoveryRSI

An open research framework and 27B model that recursively develop and evaluate scientific hypotheses, focused on life-science research.

View on GitHubWebsite
Stars
107
Forks
8
Open issues
0
Last commit
11 hours ago

What is it?

DiscoveryRSI treats scientific research as a recursive loop centered on hypotheses. Each proposal is a versioned research object that combines an evidence-grounded claim, an investigation plan, explicit decision criteria, and alternative explanations. Multi-agent search develops these hypotheses into a research graph, and selected computational investigations are compiled into executable Harbor tasks. The evidence and revisions behind each step are stored for reuse.

The project includes DiscoveryRSI-27B, a 27B model trained with Hypothesis Credit Optimization, which links assessments of scientific quality to specific contributions such as adding evidence or a comparator. The public harness provides the agentic core, provider adapters for OpenAI-compatible endpoints, role prompts, a tool registry, a replay store, and campaign manifests, and it can be run with a local demo or a served model via vLLM.

The companion DFM-Bench evaluates research progress at the level of individual hypotheses, revisions, full research runs, and deduplicated contributions, with a life-science setting and a battery-science extension. Outputs are intended for expert review and independent verification, not clinical decision-making.

Key features

  • 01Generates versioned hypotheses with evidence, investigation plans, and decision criteria.
  • 02Tracks parent links and revision lineage across research iterations.
  • 03Compiles selected computational investigations into executable Harbor tasks.
  • 04Provides a DiscoveryRSI-27B model served through an OpenAI-compatible endpoint.
  • 05Includes a local demo, tests, and replay storage for campaign runs.
  • 06Pairs with DFM-Bench for evaluating cumulative research progress.

Use cases

  • Generating and refining testable hypotheses in life-science research.
  • Prioritizing therapeutic targets or exploring disease mechanisms with authorized data.
  • Studying how AI research agents develop ideas over multiple iterations.
  • Benchmarking research-agent systems at the level of full research paths.

Quick start

  1. 1

    Clone the repository

    git clone https://github.com/Gen-Verse/DiscoveryRSI.git
  2. 2

    Enter the project directory

    cd DiscoveryRSI
  3. 3

    Install the public harness

    python -m pip install -e .
  4. 4

    Run the deterministic local demo

    discovery-rsi demo --iterations 2

Things to know

  • Serving the 27B model for the full campaign workflow requires self-hosting an OpenAI-compatible endpoint, such as with vLLM.
  • Generated hypotheses can be plausible but unsupported and require expert review and independent verification.
  • The system is not a medical device and must not be used for clinical or other high-stakes decisions.
  • Only data you are authorized to use can be brought in, since no patient or restricted datasets are bundled.

Topics

#ai-for-science#llm-agents#agent-framework#hypothesis-generation#life-science#benchmark#research-automation#agentic-workflow#aiforscience#recursive-self-improvement

FAQ

What is DiscoveryRSI?
An open research framework and 27B model that recursively develop and evaluate scientific hypotheses, focused on life-science research. DiscoveryRSI treats scientific research as a recursive loop centered on hypotheses. Each proposal is a versioned research object that combines an evidence-grounded claim, an investigation plan, explicit decision criteria, and alternative explanations. Multi-agent search develops these hypotheses into a research graph, and selected computational investigations are compiled into executable Harbor tasks. The evidence and revisions behind each step are stored for reuse.
Is DiscoveryRSI free to use? What license does it have?
DiscoveryRSI is open source under the Other license. Check the license terms before using it in commercial products.
Is DiscoveryRSI still maintained?
Actively maintained. The last commit to DiscoveryRSI was on Oct 20, 2026.
What can I use DiscoveryRSI for?
Generating and refining testable hypotheses in life-science research. Prioritizing therapeutic targets or exploring disease mechanisms with authorized data. Studying how AI research agents develop ideas over multiple iterations. Benchmarking research-agent systems at the level of full research paths.
How do I install DiscoveryRSI?
Clone the repository: git clone https://github.com/Gen-Verse/DiscoveryRSI.git Enter the project directory: cd DiscoveryRSI Install the public harness: python -m pip install -e . Run the deterministic local demo: discovery-rsi demo --iterations 2
Where can I get DiscoveryRSI?
The source code and installation instructions are on GitHub: https://github.com/Gen-Verse/DiscoveryRSI

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects