R&D-Agent
A framework that automates high-value research-and-development work in AI, letting AI agents drive the model and data side of AI research, from ideation through implementation to evaluation.
R&D-Agent is a framework from Microsoft Research for automating research-and-development processes in AI, letting AI systems drive the data- and model-centric work of building other AI systems. It provides a general data-science agent, a Kaggle-competition agent, and specialized agents for quantitative trading research.
The project reports that R&D-Agent is the top-performing machine learning engineering agent on MLE-bench, a benchmark built from 75 Kaggle competitions that assesses AI systems on realistic machine-learning engineering tasks; related work includes a benchmark for whether LLM agents can autonomously engineer end-to-end post-training pipelines spanning static optimization and stateful reinforcement learning.
It now fully supports LiteLLM as its default backend, so it can be pointed at multiple LLM providers, and ships a web frontend, built with a server command, for real-time interaction and trace viewing.