RAG Chatbot w/ coHere ReRanker

RAG Chatbot with Data Ingestion & Sync This workflow creates a comprehensive, automated Retrieval-Augmented Generation (RAG) system. It builds and maintains a knowledge base from documents in a Google Drive folder, and exposes that knowledge through an intelligent AI chatbot that can cite its sources. The system is divided into three distinct parts: the user-facing chatbot, an automated data ingestion pipeline, and a cleanup process to keep the data synchronized. Technologies Used - Orchestration: n8n - AI Language Models: OpenAI (GPT-4.1-mini for chat, text-embedding-3-small for embeddings) - AI Reranking: Cohere - Data Source: Google Drive - Vector Database: Supabase (using Postgres with the pgvector extension) - Chat Memory: Postgres Automation & Value Proposition This workflow provides an end-to-end solution for turning a folder of documents into an interactive, accurate, and self-maintaining AI assistant. - Automated Knowledge Base Management: It fully automates the process of monitoring a Google Drive folder, extracting text from new or updated documents (PDFs, Google Docs), and populating a vector database. This eliminates the manual effort of keeping a knowledge base current. - High-Quality, Citable Answers: By using a sophisticated RAG pattern with a reranker, the chatbot provides answers that are not only relevant but also grounded in the provided documents. The requirement to cite sources builds user trust and allows for verification. - Data Integrity: The automated cleanup routine prevents the knowledge base from becoming stale. It periodically checks for documents that have been deleted from Google Drive and removes the corresponding data from the vector store, ensuring the chatbot always relies on the most current information. - Persistent Memory: The chatbot remembers previous interactions within a session, allowing for more natural, contextual follow-up conversations. AI Integration AI is deeply integrated at every critical stage of this workflow, creating a sophisticated multi-step reasoning and data processing pipeline. 1.