DeepYard
R

Replica

Auto-evaluating framework for training AI agents to replicate scientific research papers

Open SourceFree

About

Replica is a research framework that trains AI agents to reproduce scientific papers through hypothesis-driven exploration. It features a scalable task space where agents learn to formulate hypotheses, design experiments, and validate findings against original papers. The system uses auto-generated rubrics to evaluate agent performance across replication tasks, enabling systematic development of research-capable AI agents. Designed for AI safety researchers and those building autonomous scientific discovery systems.

Details

Type
Integrations
Language

Tags

autonomousevaluationopen-sourceframeworkpython