Aris Autonomous Ml Research

This is a complete autonomous ML research system built entirely from Markdown files that any LLM can read and execute. The core idea is smart: one model (Claude Code or Codex) runs your research while a different model (GPT, Gemini, MiniMax) acts as adversarial reviewer to catch blind spots. It handles the full pipeline from paper review to idea generation to experiment automation to LaTeX paper writing, all designed to run overnight. The cross-model review loop is the real trick here, iterating until papers hit "clear accept" scores. Setup requires copying Markdown skills to your Claude Code directory and configuring MCP servers for the reviewer model. It includes rebuttal generation with character limits and safety gates against fabrication, plus bonus tools for slides and posters. Zero dependencies beyond the LLM agents themselves.