Gate AI Answers w/ Groq Judge & Source Docs in n8n

Gate AI Answers with a Groq Judge + Source Docs in n8n

This n8n workflow automatically checks whether an AI-generated answer is supported by the provided source text—using deterministic safeguards and a Groq LLM judge—then returns a clear PASS / REVIEW / BLOCK verdict and logs results to an n8n data table.

What this workflow does

• Starts from either a sub-workflow call (inputs: answer and sources) or the workflow’s built-in n8n form.
• Applies a configurable gate policy, including review/block thresholds, a model score cap, and a minimum source length (it refuses to PASS if sources are too short).
• Runs deterministic checks to detect numbers and URLs present in the answer but missing from the provided sources, assigning risk points.
• Judges unsupported claims with Groq: sends the answer + sources to a selected Groq chat model to extract unsupported claims, each with a short reason and a confidence score.
• Computes a final risk score by combining deterministic findings and capped Groq-based points, then outputs a PASS / REVIEW / BLOCK verdict.
• Logs everything (verdict, score, findings) into an n8n data table, and either shows a results page (form submissions) or returns the verdict details to the calling workflow.

Use cases

• Route AI responses in a SaaS support agent so answers without sufficient source backing are flagged before sending.
• Build a reliable RAG quality gate: require that generated claims (including URLs/numbers) appear in the supplied documents.
• Let automation engineers audit model behavior by storing pass/review/block outcomes and reasons in a data table.

Technical details

• Groq integration: add a Groq API credential and select the Groq chat model used as the judge.
• n8n nodes used: if, set, code, form, data table, sticky note.
• Logging: configure the Data Table ID for the check log.