microsoft/rd-agent automates high-value, data-driven industrial R&D processes by having AI autonomously extract formulas and models from real-world materials, implement them, and iteratively propose new ideas. It serves as an evolving framework that learns from feedback and knowledge to continuously improve R&D capabilities.
Project overview
It is recognized as a top-performing machine learning engineering agent on MLE-bench and is the first data-centric quantitative multi-agent framework.
Project type
AI Agent · AI Coding · Model Development · Data Processing
Use cases
Data Analysis · Automation
Deployment
Refer to project documentation
License
MIT
Best for
Data teams and AI engineers who need to automate data-driven R&D processes including model and factor development.
Researchers who need to extract and implement model structures from papers and financial reports.
Key capabilities
Automates quantitative trading via self-loop factor and model proposal, extraction, and joint implementation using Qlib.
Automates model tuning and feature engineering for Kaggle and general data science competitions.
Automatically reads research papers and financial reports to extract and implement model structures and datasets.
Autonomous LLM fine-tuning for benchmark-driven domain adaptation, including data processing, training, and evaluation.
Automates medical self-loop model proposal and implementation.
Framework that learns from feedback and knowledge to continuously improve R&D capabilities like human experts.
Limitations and risks
The project currently only supports Linux.
The current web UI does not support the data_science scenario yet.
Docker must be installed before attempting most scenarios.
Executing scenarios requires external LLM APIs which incur monetary costs.
Getting started
Docker must be installed first for most scenarios. Create a conda environment with Python 3.10.
Install the package with pip (pip install rdagent), then configure a .env file with the required external LLM API keys.
Run rdagent health_check to verify the configuration, then launch a scenario such as rdagent fin_factor.
Evidence and sources
GitHub project description: We are committed to automating these high-value generic R&D processes through R&D-Agent, which lets AI drive data-driven AI.
README: we are aiming to build an Agent to automate Data-Driven R\&D that can + 📄 Read real-world material (reports, papers, etc.) and **extract** key formulas, descriptions of interested…