Basic RAG Chatbot for Document-Based Sales Queries

This n8n workflow implements a basic RAG pipeline divided into two main sections. Part 1 focuses on data ingestion: it reads files from disk or Google Drive, splits them into chunks using a Recursive Character Text Splitter, generates embeddings via the Cohere API, and stores them in an In-Memory Vector Store (easily swappable for Pinecone or Qdrant). This setup allows for efficient indexing of sales documents, CRM data, or knowledge bases.

Part 2 handles interactive querying: user input from a chat trigger is embedded with the same Cohere model, relevant chunks are retrieved via similarity search, and a Groq-hosted LLM generates accurate, context-grounded responses. The workflow leverages n8n's visual orchestration for easy debugging and scaling.

Benefits include reduced hallucinations in AI responses, fast prototyping for sales chatbots, and seamless integration into CRM systems for customer support or lead qualification. Use cases span sales teams querying product docs, CRM knowledge retrieval, or internal Q&A bots, saving hours on manual research and boosting response accuracy.