¡En Vivo Contigo!
Budget / Salary$250–750
TypeFreelance project
LocationRemote
Posted39 minutes ago
I want a complete n8n workflow that funnels every website-chat message through two AI agents—Anthropic Claude and Google Gemini—so visitors get instant, accurate answers about their own account, order, or booking.
Here’s how I picture the flow: a webhook catches the chat event, n8n verifies the customer’s identity, pulls the relevant account or order details from our database/CRM, then feeds that context to Claude. If Claude’s confidence score drops below an agreed threshold or a policy rule is triggered, Gemini steps in as either a fallback or a second opinion. The final answer is returned to the chat widget in real time, all within a few seconds.
Key deliverables
• Production-ready n8n workflow file (.json) with clear node naming and comments
• Secure connectors to our database/CRM (REST or SQL, whichever you choose) that surface only the fields required for support answers
• Claude and Gemini prompt templates that inject the retrieved customer data and enforce tone/formatting guidelines
• Error-handling, retry, and logging nodes so failed calls are flagged and stored for review
• Environment variable setup and a quick-start README so I can import the flow, drop in my API keys, and go live
• Demo recording or live walkthrough proving the agent answers typical chat questions—“Where’s my order?”, “What’s the status of booking #1234?”, “Can I change my delivery address?”—with no manual intervention
Acceptance criteria
1. A visitor receives a correct, personalised reply in under 5 seconds.
2. Sensitive data never appears in logs or model prompts unless explicitly required for the answer.
3. Switching model order (Gemini first, Claude fallback) is a one-line change in the workflow.
If you’ve built multi-model automations in n8n before—or have clever ideas on prompt engineering, vector stores, or function calling to sharpen responses—I’m all ears. Let’s ship this fast.
Here’s how I picture the flow: a webhook catches the chat event, n8n verifies the customer’s identity, pulls the relevant account or order details from our database/CRM, then feeds that context to Claude. If Claude’s confidence score drops below an agreed threshold or a policy rule is triggered, Gemini steps in as either a fallback or a second opinion. The final answer is returned to the chat widget in real time, all within a few seconds.
Key deliverables
• Production-ready n8n workflow file (.json) with clear node naming and comments
• Secure connectors to our database/CRM (REST or SQL, whichever you choose) that surface only the fields required for support answers
• Claude and Gemini prompt templates that inject the retrieved customer data and enforce tone/formatting guidelines
• Error-handling, retry, and logging nodes so failed calls are flagged and stored for review
• Environment variable setup and a quick-start README so I can import the flow, drop in my API keys, and go live
• Demo recording or live walkthrough proving the agent answers typical chat questions—“Where’s my order?”, “What’s the status of booking #1234?”, “Can I change my delivery address?”—with no manual intervention
Acceptance criteria
1. A visitor receives a correct, personalised reply in under 5 seconds.
2. Sensitive data never appears in logs or model prompts unless explicitly required for the answer.
3. Switching model order (Gemini first, Claude fallback) is a one-line change in the workflow.
If you’ve built multi-model automations in n8n before—or have clever ideas on prompt engineering, vector stores, or function calling to sharpen responses—I’m all ears. Let’s ship this fast.
Apply on Freelancer →
Project sourced from Freelancer.com. Applications happen directly on the original platform — we never collect your data.