💬 Dual-path customer support system with Google Sheets, vectors & Gemini
⚡ 238 views · 💬 Support Chatbots
Description
This n8n workflow template implements a dual-path architecture for AI customer support, based on the principles outlined in the research paper “A Locally Executable AI System for Improving Preoperative Patient Communication: A Multi-Domain Clinical Evaluation” (Sato et al.).
The system, named LENOHA (Low Energy, No Hallucination, Leave No One Behind Architecture), uses a high-precision classifier to differentiate between high-stakes queries and casual conversation. Queries matching a known FAQ are answered with a pre-approved, verbatim response, structurally eliminating hallucination risk. All other queries are routed to a standard generative LLM for conversational flexibility.
This template provides a practical ++blueprint++ for building safer, more reliable, and cost-efficient AI agents, particularly in regulated or high-stakes domains where factual accuracy is critical.
What This Template Does (Step-by-Step)
- Loads an expert-curated FAQ from Google Sheets and creates a searchable vector store from the questions during a one-time setup flow.
- Receives incoming user queries in real-time via a chat trigger.
- Classifies user intent by converting the query to an embedding and searching the vector store for the most semantically similar FAQ question.
- Routes the query down one of two paths based on a configurable similarity score threshold.
- Responds with a verbatim, pre-approved answer if a match is found (safe path), or generates a conversational reply via an LLM if no match is found (casual path).
Important Note for Production Use
This template uses an in-memory Simple Vector Store for demonstration purposes. For a production application, this should be replaced with a persistent vector database (e.g., Pinecone, Chroma, Weaviate, Supabase) to store your embeddings permanently.
Required Integrations:
- Google Sheets (for the FAQ knowledge base)
- Hugging Face API (for creating embeddings)
- An LLM provider (e.g., OpenAI, Anthropic, Mistral)
- (Recommended) A persistent Vector Store integration.
Best For:
🏦 Organizations in regulated industries (finance, healthcare) requiring high accuracy. 💰 Applications where reducing LLM operational costs is a priority. ⚙️ Technical support agents that must provide precise, unchanging information. 🔒 Systems where auditability and deterministic responses for known issues are required.
Key Benefits:
✅ Structurally eliminates hallucination risk for known topics. ✅ Reduces reliance on expensive generative models for common queries. ✅ Ensures deterministic, accurate, and consistent answers for your FAQ. ✅ Provides high-speed classification via vector search. ✅ Implements a research-backed architecture for building safer AI systems.
🔗 Nodes Used
Google Sheets, Embeddings Hugging Face Inference, Simple Vector Store, Default Data Loader, Chat Trigger, Google Gemini
📥 Import
Download workflow.json and import into n8n:
Workflow menu → Import from File