Repolode
Submit
microsoft

microsoft/RD-Agent

An LLM agent framework from Microsoft that automates data-driven R&D tasks such as reading papers, implementing models and factors, and running Kaggle-style experiments.

View on GitHubWebsite
Stars
14,993
Forks
2,142
Open issues
251
Last commit
4 days ago

What is it?

R&D-Agent is a Python framework from Microsoft that uses LLM agents to automate data-centric research and development. It splits the work into an R part that proposes ideas and a D part that implements them as runnable code, then iterates using execution feedback.

The project ships several scenarios: quantitative finance (generating factors and models on Qlib), extracting formulas from research papers or financial reports, medical prediction model evolution, Kaggle-style model tuning and feature engineering, and autonomous LLM fine-tuning (FT-Agent). It also includes a benchmark for LLM agents on post-training engineering (Agent² RL-Bench).

Setup requires Linux, Docker, and Python 3.10 or 3.11. Models are accessed through LiteLLM, so chat and embedding providers can be configured via environment variables. Monitoring is available through a Streamlit UI or a Flask-based web UI.

Key features

  • 01Proposes research ideas and implements them as runnable code in an iterative loop.
  • 02Extracts formulas and model structures from papers and financial reports.
  • 03Runs quantitative factor and model optimization on Qlib.
  • 04Supports Kaggle-style model tuning and feature engineering.
  • 05Connects to multiple LLM providers through LiteLLM.
  • 06Provides Streamlit and Flask web UIs for viewing run traces.

Use cases

  • Automating factor mining and model iteration for quantitative trading research.
  • Reproducing model architectures from research papers as code.
  • Running automated baseline experiments for Kaggle-style competitions.
  • Fine-tuning LLMs on benchmark-driven tasks with automated evaluation.

Quick start

  1. 1

    Ensure Docker is installed and runnable without sudo

  2. 2

    Create a conda environment with Python 3.10

    conda create -n rdagent python=3.10
  3. 3

    Activate the conda environment

    conda activate rdagent
  4. 4

    Install R&D-Agent from PyPI

    pip install rdagent

Things to know

  • RD-Agent currently only supports Linux.
  • Most scenarios require Docker and an LLM API configuration.

Topics

#agent#llm#automation#data-science#machine-learning#research#quantitative-finance#ai#data-mining#development

FAQ

What is RD-Agent?
An LLM agent framework from Microsoft that automates data-driven R&D tasks such as reading papers, implementing models and factors, and running Kaggle-style experiments. R&D-Agent is a Python framework from Microsoft that uses LLM agents to automate data-centric research and development. It splits the work into an R part that proposes ideas and a D part that implements them as runnable code, then iterates using execution feedback.
Is RD-Agent free to use? What license does it have?
RD-Agent is open source under the MIT license. Check the license terms before using it in commercial products.
Is RD-Agent still maintained?
Actively maintained. The last commit to RD-Agent was on Oct 16, 2026.
What can I use RD-Agent for?
Automating factor mining and model iteration for quantitative trading research. Reproducing model architectures from research papers as code. Running automated baseline experiments for Kaggle-style competitions. Fine-tuning LLMs on benchmark-driven tasks with automated evaluation.
How do I install RD-Agent?
Ensure Docker is installed and runnable without sudo Create a conda environment with Python 3.10: conda create -n rdagent python=3.10 Activate the conda environment: conda activate rdagent Install R&D-Agent from PyPI: pip install rdagent
Where can I get RD-Agent?
The source code and installation instructions are on GitHub: https://github.com/microsoft/RD-Agent

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

Related projects