
scholar-loop
renee-jia
An autonomous AI scientist: a multi-agent loop over literature, experiments, self-critique and write-up, with deterministic guards against reward-hacking and hallucination.
AI 简介
Scholar Loop 是一个面向科研自动化的多智能体系统,模拟博士生完整研究流程:文献调研、假设生成、真实机器学习实验(PyTorch)、结果反思与论文撰写。其核心特点是采用确定性防护机制(deterministic guards)防止大模型幻觉与奖励作弊,通过8个角色分明的智能体协同工作,并内置预算感知的三级实验筛选漏斗(smoke→verify→full)和可插拔的实验引擎架构。适用于小规模AI科研探索、教学演示、自动化实验验证及LLM驱动的研究方法学研究。
Python
MIT License461
Stars
35
Forks
30
Watchers
0
Issues
Star 增长
今日0
近 7 天0
近 30 天+113
综合评分54.67
默认分支main