AI-Powered WhatsApp RAG Chatbot with Multimedia Support

Unlock Intelligent Customer & Internal Support with this AI Agent

This AI Agent transforms your WhatsApp channel into a smart, multilingual support hub. It consists of two n8n workflows: one for ingesting and indexing your knowledge base (e.g., product manuals from Google Docs), and another for handling live WhatsApp conversations. The agent uses OpenAI's GPT-4o-mini and embedding models, along with MongoDB Atlas for vector storage, to deliver accurate responses based on your specific documentation. It supports various input types including text, voice (transcribed), images (analyzed), and PDF documents, making interactions natural and efficient.

Key Features & Benefits

• Knowledge Base Automation: Automatically ingests documents from Google Docs, chunks them, generates embeddings (OpenAI), and stores them in MongoDB Atlas for fast semantic search.
• Multimedia WhatsApp Interaction: Responds to user queries via text, transcribed voice notes, analyzed images, and parsed documents, offering a versatile communication experience.
• AI-Powered RAG: Leverages Retrieval Augmented Generation with OpenAI's GPT-4o-mini to provide accurate, contextually relevant answers drawn directly from your indexed knowledge.
• Conversational Memory: Maintains conversation context across multiple turns using a memory buffer, ensuring coherent and natural dialogues.
• Efficient Support: Frees up human agents by handling common queries, providing instant responses, and improving overall support team productivity.
• Customizable AI Persona: Tailor the AI's tone, answering style, and business rules through a configurable system prompt.

• An n8n instance (Cloud or self-hosted).
• OpenAI API Key with access to GPT-4o-mini (or a comparable model) and text embedding models (e.g., text-embedding-ada-002).
• MongoDB Atlas account with a database and collection. You'll need to create a vector search index on this collection.
• Google Docs credentials (OAuth2) for the document ingestion workflow.
• WhatsApp Business API access via a Meta Developer Account and a configured App.

1. Download the n8n workflow JSON file. This file contains two distinct workflows: 'Document Ingestion & Indexing' and 'AI-Powered Query & Response via WhatsApp'.
2. Import both workflows into your n8n instance.

Workflow 1: Document Ingestion & Indexing Setup 3. Open the 'Document Ingestion & Indexing' workflow. 4. Google Docs Node: Configure with your Google Docs credentials and specify the URL of the document(s) containing your product documentation/knowledge base. 5. Embeddings OpenAI Node: Enter your OpenAI API Key and select an appropriate embedding model. 6. MongoDB Atlas Vector Store Node (Upsert Documents operation): * Authenticate with your MongoDB Atlas credentials. * Specify your database, collection name, the field for text content (e.g., text), and the field where embeddings will be stored (e.g., embedding). * Define the vector search index name (e.g., data_index). Ensure this index is created in MongoDB Atlas with the correct dimensions for your chosen embedding model (e.g., 1536 for text-embedding-ada-002). Refer to the example search index in the original workflow description. 7. Run this workflow manually once to ingest and index your documentation. Schedule or re-run it as needed when your documentation updates.

Workflow 2: AI-Powered Query & Response via WhatsApp Setup 8. Open the 'AI-Powered Query & Response via WhatsApp' workflow. 9. WhatsApp Trigger Node: Configure with your WhatsApp Business API credentials from your Meta Developer Account. 10. MongoDB Atlas Vector Store Node (Similarity Search operation): * Authenticate with your MongoDB Atlas credentials (same as in Workflow 1). * Connect to the same database, collection, and specify the same vector search index name (e.g., data_index) and embedding field used in the ingestion workflow. 11. AI Agent Node (or relevant OpenAI Chat Model Node): * Enter your OpenAI API Key. * Select the chat model (e.g., gpt-4o-mini). * Customize the System Message to define the AI's persona, tone, answering style, and any specific business rules. Crucially, instruct it to use the context retrieved from the MongoDB vector search. 12. (If applicable) HTTP Request/OpenAI Nodes for Multimedia: Review nodes responsible for voice transcription (e.g., OpenAI Whisper) or image analysis. Ensure API keys and configurations are correct if these use separate services or specific OpenAI functionalities. 13. Simple Memory Node: Adjust settings if needed to control the conversation history window. 14. Final Check: Ensure both MongoDB nodes across the two workflows are configured to use the exact same collection, embedding field name, and vector index name. 15. Activate the 'AI-Powered Query & Response via WhatsApp' workflow.