n8n AI Support Automation
Budget / Salary$250–750
TypeFreelance project
LocationRemote
Posted1 hour ago
I’m building an email-only support desk that runs entirely on n8n and taps two large-language-model endpoints—Claude and Gemini—so that every incoming message in our inbox is answered automatically. The agent’s single purpose is to handle customer questions and support requests with the best possible answer it can assemble from those two models. When the combined confidence drops below an agreed threshold, the workflow must route the original email (with the draft answer and model reasoning attached) straight to a human teammate for manual follow-up.
Here’s what I need from you:
• A fully commented n8n workflow (exportable JSON) that watches the inbox via IMAP/Gmail, invokes Claude and Gemini sequentially or in parallel, chooses or blends their responses, then sends the reply through SMTP/Gmail.
• Clear logic for confidence scoring and the automatic escalation path to a human address or shared mailbox when the answer isn’t reliable.
• Prompt templates, environment variable names, and any helper functions written in Code nodes so I can fine-tune behaviour later.
• Short setup notes covering keys, quotas, and the exact model versions used, plus a test plan so I can verify that (a) confident questions receive an immediate answer and (b) uncertain questions are forwarded correctly.
If you’re comfortable chaining multi-model calls in n8n and know how to keep email threads intact, I’d love to see how you’d tackle this.
Here’s what I need from you:
• A fully commented n8n workflow (exportable JSON) that watches the inbox via IMAP/Gmail, invokes Claude and Gemini sequentially or in parallel, chooses or blends their responses, then sends the reply through SMTP/Gmail.
• Clear logic for confidence scoring and the automatic escalation path to a human address or shared mailbox when the answer isn’t reliable.
• Prompt templates, environment variable names, and any helper functions written in Code nodes so I can fine-tune behaviour later.
• Short setup notes covering keys, quotas, and the exact model versions used, plus a test plan so I can verify that (a) confident questions receive an immediate answer and (b) uncertain questions are forwarded correctly.
If you’re comfortable chaining multi-model calls in n8n and know how to keep email threads intact, I’d love to see how you’d tackle this.
Apply on Freelancer →
Project sourced from Freelancer.com. Applications happen directly on the original platform — we never collect your data.