Back to browse
GitHub Repository

A multi-agent protocol pairing a tool-using Scientist with a question-only advisor — no tools, no answers, no directives — improves Kaggle test performance on 4 of 5 MLE-bench tasks

3 starsPython

Multi Agent Protocol for AI Scientist by Hexo Labs

by martianvoid·Jun 25, 2026·2 points·0 comments

AI Analysis

●●●BangerBig BrainZero to One

Question-only advisor constraint unlocks 55.9% better Kaggle scores — genuinely novel agent architecture.

Strengths
  • Advisor can never give answers or directives — only ask questions and approve plans.
  • Measurable 55.9% average improvement across five MLE-bench Kaggle competitions.
  • Stateful advisor paired with stateless scientist creates effective coordination.
Weaknesses
  • Requires both Anthropic and OpenAI API keys for full functionality.
  • GPU needed for deep learning tasks like Statoil and NFL benchmarks.
Category
Target Audience

AI researchers, ML engineers, agent developers

Similar To

CrewAI · AutoGen · LangGraph

Similar Projects

Developer Tools●●●Banger

Boardroom MCP - Multi-advisor governance engine for AI agents

Multi-advisor debate engine forces tension in agent decisions, not just chaining prompts.

Big BrainZero to One
rsalars
314mo ago