Skip to content
Every project

Multi-Agent Arbitration

Three specialist critics on three different providers, resolving to one confidence-scored verdict.

What it does

Asking one LLM to grade its own output produces a biased answer. Here three critics on OpenAI, Anthropic and Ollama grade along distinct dimensions, a pure disagreement detector measures where they conflict, and an adjudicator agent synthesises the final verdict. LangGraph-style fixed DAG with a deterministic fallback path.

Architecture

outputaccuracyconsistencysafetyconflictadjudicator