WhatsApp Sales Agent with Product Brochure Knowledgebase

• WhatsApp Integration: Real-time message handling via the WhatsApp Trigger and reply via the WhatsApp node.
• AI-Powered Sales Agent: Utilizes OpenAI's GPT-4o for intelligent conversation and response generation.
• Dynamic Knowledgebase: Creates an in-memory vector store from a product brochure PDF for context-aware responses.
• Document Processing: Extracts text content from PDF files using the 'Extract from File' node.
• Text Splitting: Prepares document text for embedding using the 'Recursive Character Text Splitter'.
• Vector Store Management: Uses 'VectorStoreInMemory' for creating and querying the knowledgebase.
• Conversation Memory: 'WindowBufferMemory' ensures the AI agent can recall past interactions within a session.
• Message Type Handling: Differentiates between text and non-text messages, responding accordingly.
• Tool Usage: Integrates a 'VectorStoreTool' to allow the AI agent to query the product brochure.

Part 1: Setting Up the Product Knowledgebase (Run Once)

1. Download Product Brochure: The get Product Brochure node fetches the 'Yamaha-Powered-Loudspeakers-brochure-2024-en-web.pdf' from the provided URL.
2. Extract Text: The Extract from File node reads the downloaded PDF and extracts its text content.
3. Embeddings: The Embeddings OpenAI1 node generates embeddings for the extracted text.
4. Create Vector Store: The Create Product Catalogue node uses the embeddings and text to build an in-memory vector store. This node is configured to clear the store and insert new data, effectively creating or updating the knowledgebase.
5. Data Loading and Splitting: The Default Data Loader and Recursive Character Text Splitter nodes prepare the document data for the vector store.

Note: For production, consider using persistent vector stores like Qdrant or Pinecone instead of 'VectorStoreInMemory'.

Part 2: WhatsApp AI Chatbot (Activated Workflow)

1. WhatsApp Trigger: The WhatsApp Trigger node listens for incoming messages on your WhatsApp account.
2. Message Type Handling: The Handle Message Types switch node checks if the incoming message is of type 'text'.
   • If 'text', it proceeds to the AI Sales Agent.
   • If not 'text', it sends a predefined message via Reply To User1 informing the user that only text messages are supported.
3. AI Sales Agent: The AI Sales Agent node takes the incoming text message and uses the configured OpenAI model (OpenAI Chat Model) and system message. It also receives memory from Window Buffer Memory and can access tools, specifically the Vector Store Tool (69f1b78b-7c93-4713-863a-27e04809996f) which is linked to the Product Catalogue vector store.
4. Response Generation: The AI agent processes the query against the product brochure data.
5. Reply to User: The Reply To User node sends the AI agent's generated response back to the customer via WhatsApp.

Activation: Ensure your workflow is activated to start receiving and responding to WhatsApp messages. For self-hosted n8n instances, make sure your server is accessible from WhatsApp.

Language: The AI model gpt-4o-2024-08-06 is capable of understanding and responding in multiple languages, including Chinese. When a Chinese query is received, the AI model will process it and generate a response in Chinese, utilizing the English product brochure content as its knowledge base. The workflow itself does not explicitly translate the brochure content but relies on the AI's multi-lingual capabilities.