n8n AI Chatbot: Web & WhatsApp
Budget / Salary£250–750
TypeFreelance project
LocationRemote
Posted2 hours ago
I’m building an AI-powered support assistant orchestrated entirely in n8n. In this first phase the bot must respond on two fronts—our website’s live-chat widget and our WhatsApp number—using the same brain and conversation history regardless of channel.
Knowledge base
The answers should be generated from:
• Content scraped or pulled via API from our public help-center website
• A collection of PDFs/manuals I will upload
Retrieval, embedding, and prompt-engineering logic can live inside n8n itself or call out to external services (e.g. OpenAI, Pinecone, Supabase, Weaviate); I’m open as long as the workflow remains transparent and editable in n8n.
Fallback flow
Whenever the model’s confidence drops below an agreed threshold it must hand the dialogue—along with full context—to a human agent. I don’t want the bot asking customers for extra details; it should immediately escalate.
Deliverables
1. n8n workflow (export file) covering: ingestion of sources, vector store updates, real-time chat handling for website and WhatsApp, fallback routing.
2. Installation/connection notes for the live-chat widget you choose (if it’s not already native to n8n) and the WhatsApp API (Twilio, 360dialog, or other).
3. Short Loom or markdown walkthrough so my team can extend the flow and add new documents later.
Acceptance criteria
• Bot answers at least 80 % of a supplied test set correctly within 2 seconds.
• Low-confidence queries are routed to the human queue every time, with transcript attached.
• Entire solution runs in my self-hosted n8n instance with no proprietary vendor lock-in.
If this sounds like a build you can knock out efficiently, let’s talk timelines and any prerequisites you have.
Knowledge base
The answers should be generated from:
• Content scraped or pulled via API from our public help-center website
• A collection of PDFs/manuals I will upload
Retrieval, embedding, and prompt-engineering logic can live inside n8n itself or call out to external services (e.g. OpenAI, Pinecone, Supabase, Weaviate); I’m open as long as the workflow remains transparent and editable in n8n.
Fallback flow
Whenever the model’s confidence drops below an agreed threshold it must hand the dialogue—along with full context—to a human agent. I don’t want the bot asking customers for extra details; it should immediately escalate.
Deliverables
1. n8n workflow (export file) covering: ingestion of sources, vector store updates, real-time chat handling for website and WhatsApp, fallback routing.
2. Installation/connection notes for the live-chat widget you choose (if it’s not already native to n8n) and the WhatsApp API (Twilio, 360dialog, or other).
3. Short Loom or markdown walkthrough so my team can extend the flow and add new documents later.
Acceptance criteria
• Bot answers at least 80 % of a supplied test set correctly within 2 seconds.
• Low-confidence queries are routed to the human queue every time, with transcript attached.
• Entire solution runs in my self-hosted n8n instance with no proprietary vendor lock-in.
If this sounds like a build you can knock out efficiently, let’s talk timelines and any prerequisites you have.
Apply on Freelancer →
Project sourced from Freelancer.com. Applications happen directly on the original platform — we never collect your data.