n8n LLM Arbitration: Claude Panel Consensus + WhatsApp Alerts

Get consensus answers from multiple Claude models—then alert your team on low-confidence cases

This n8n LLM Arbitration workflow coordinates a panel of Anthropic Claude models, measures weighted agreement using a Claude adjudicator, and escalates uncertain results to an arbiter—with WhatsApp alerts when human review may be needed.

What this workflow does

• Receives a request via POST webhook at /webhook/llm-arbitration (or runs manually), normalizing input into a question, context, criteria, and arbitration settings.
• Validates input: if the question is empty, it returns a 400 JSON error response to the webhook.
• Builds a Claude panel and queries multiple Anthropic Claude models in parallel to produce structured outputs (including key claims and a confidence score).
• Adjudicates disagreement: a separate Claude adjudicator clusters equivalent positions and extracts the main points of disagreement.
• Computes weighted agreement from panelist weights to decide whether to return a consensus answer, run another debate round, or escalate to arbitration when the round limit is reached.
• Runs arbitration when needed: calls an Anthropic Claude arbiter (optionally with web search enabled) to issue a verdict with rationale, dissent, verified claims, and a human-review flag.
• Responds with the final result to the original webhook, including per-round agreement and the arbitration outcome.

Use cases

• Reduce hallucinations for customer support and knowledge-base Q&A by requiring panel consensus.
• Automate policy, compliance, or contract interpretation with debate rounds and escalation.
• Route low-confidence LLM answers to an arbiter and notify teams via WhatsApp.

Technical details

• n8n nodes/logic: webhook, if, set, code, switch, and WhatsApp.
• Anthropic Claude integration for panelists, adjudicator, and arbiter (with optional web search).