WhatsApp Sales Assistant with Product Brochure Knowledgebase

• WhatsApp Integration: Triggers on incoming WhatsApp messages and responds directly.
• Product Knowledgebase: Builds a vector store from a PDF product brochure for efficient information retrieval.
• AI Agent: Employs an AI agent (powered by OpenAI) to understand user queries and generate responses.
• Contextual Memory: Maintains conversation history per user using window buffer memory.
• Message Type Handling: Filters out non-text messages and provides a polite refusal.
• Scalable: Designed for potential expansion with production-ready vector stores like Qdrant or Pinecone.

Workflow Setup:

1. Download Product Brochure PDF: The get Product Brochure node (HTTP Request) downloads the specified product brochure. The Extract from File node then extracts the text content from the PDF.
2. Create Product Brochure Vector Store: The extracted text is processed by Embeddings OpenAI1, Recursive Character Text Splitter, and Default Data Loader nodes. These feed into Create Product Catalogue (In-Memory Vector Store) to build the knowledge base. This step only needs to be run once initially or when the brochure is updated.
3. Use the WhatsApp Trigger: The WhatsApp Trigger node listens for incoming messages. A Handle Message Types (Switch) node filters messages, directing text messages to the AI agent and handling other types separately.
4. Sales AI Agent Responds To Customers: For text messages, the AI Sales Agent node receives the user's input. It uses OpenAI Chat Model, Window Buffer Memory for conversation context, and Vector Store Tool to query the Product Catalogue vector store. The OpenAI Chat Model1 and Vector Store Tool are configured to interact with the vector store.
5. Respond to WhatsApp User: The output from the AI Sales Agent is sent back to the user via the Reply To User (WhatsApp) node. Non-text messages receive a predefined reply from Reply To User1.