Voice AI chatbot with OpenAI, RAG (Qdrant) & Guardrails for WordPress

This workflow implements a complete Voice AI Chatbot system for Wordress that integrates speech recognition, guardrails for safety, retrieval-augmented generation (RAG), Qdrant vector search, and audio responses. It is designed to be connected to a WordPress Voicebot AI plugin through a webhook endpoint. --- Key Advantages ✅ Complete Voice AI Pipeline The workflow handles: audio input STT intelligent processing TTS output All within a single automated process. ✅ Safe and Policy-Compliant Thanks to the Guardrails module, the system automatically: detects harmful or disallowed requests blocks them responds safely This protects both the user and the business. ✅ Contextual and Memory-Based Conversations The Window Buffer Memory tied to unique session IDs enables: continuous conversation flow natural dialogue better understanding of context ✅ Company-Specific Knowledge via RAG By integrating Qdrant as a vector store, the system can: retrieve business documentation give accurate and up-to-date answers support personalized content This makes the chatbot far more powerful than a standard LLM. ✅ Modular and Extensible Architecture Because everything is modular inside n8n, you can: swap OpenAI with other models add new tools or knowledge sources change prompts or capabilities without redesigning the entire workflow. ✅ Easy WordPress Integration The workflow connects directly to a WordPress Voicebot plugin, meaning: no custom backend development simple deployment fast integration for websites ✅ Automatic Indexing of Documents The second workflow section: fetches Google Drive files converts them into embeddings indexes them into Qdrant This lets you maintain your knowledge base with almost no manual work. --- How It Works This workflow creates a Wordpress voice-enabled AI chatbot that processes audio inputs and provides contextual responses using RAG (Retrieval-Augmented Generation) from a Qdrant vector database. The system operates as follows: 1. Audio Processing Pipeline: - Receives audio input via webhook and converts speech to text using OpenAI's STT (Speech-to-Text) - Applies guardrails to detect inappropriate content or jailbreak attempts using a separate GPT-4.1-mini model - Routes safe queries to the AI agent and blocks unsafe content with a default response 2. AI Agent with Contextual Memory: - Uses OpenAI Chat Model with window buffer memory to maintain conversation context - Equips the agent with two tools: Calculator for computations and RAG tool for business knowledge retrieval - The RAG system queries Qdrant vector store containing company documents using OpenAI embeddings 3. Response Generation: - Generates appropriate text responses based on query type and available knowledge - Converts approved responses to audio using OpenAI's TTS (Text-to-Speech) with "onyx" voice - Returns binary audio responses to the webhook caller --- Set Up Steps 1. Vector Database Preparation: - Create Qdrant collection via HTTP request with specified vector configuration - Clear existing collection data before adding new documents - Set up Google Drive integration to source documents from specific folders 2. Document Processing Pipeline: - Search and retrieve files from Google Drive folder "Test Negozio" - Process documents through recursive text splitting (500 chunk size, 50 overlap) - Generate embeddings using OpenAI and store in Qdrant vector store - Implement batch processing with 5-second delays between operations 3. System Configuration: - Configure webhook endpoint for receiving audio inputs - Set up multiple OpenAI accounts for different functions (STT, TTS, guardrails, main agent) - Establish Qdrant API connections for vector storage and retrieval - Implement session-based memory management using session IDs from webhook headers 4. WordPress Integration: - Install the provided Voicebot AI Agent WordPress plugin - Configure the plugin with the webhook URL to connect to this n8n workflow - The system is now ready to receive audio queries and respond with voice answers The workflow handles both real-time voice queries and background document processing, creating a comprehensive voice assistant solution with business-specific knowledge retrieval capabilities. --- Need help customizing? Contact me for consulting and support or add me on Linkedin.