RAG AI Agent: No-Code Business Document Q&A System

Overview

This project is a no-code RAG (Retrieval-Augmented Generation) AI agent that automates document search and Q&A. Built with n8n, OpenAI, Cohere, Supabase, and Google Drive integration, it enables teams to query business documents in natural language with live responses and full source traceability.

Business Use Case

Automate knowledge management for teams by ingesting, chunking, and indexing documents from Google Drive to a secure vector database. Reduce manual document search time and enable real-time, context-aware Q&A through chat or integrated tools like Slack.

Tech Stack

• n8n (Workflow Automation)
• OpenAI (Embedding & LLM)
• Cohere (Reranking)
• Supabase (Vector Store)
• Google Drive (Document Source)
• Slack (Chat Integration, optional)

Key Features

• Automated document ingestion from Google Drive
• Intelligent document chunking and vector storage
• Semantic search with context-aware Q&A
• Source-cited answers with full metadata tracking
• Automated cleanup for deleted files
• Modular, no-code automation (powered by n8n)

Workflow

1. Trigger: Scheduled or manual workflow start.
2. Document Fetch: Syncs files from a designated Google Drive folder.
3. Processing: Splits text into overlapping chunks (750 chars), generates embeddings using OpenAI API.
4. Storage: Stores chunks and embeddings in Supabase.
5. Q&A Flow: On user question, finds relevant chunks via vector search, reranks results using Cohere, and generates a sourced answer with OpenAI.
6. Cleanup Module: Periodically identifies and removes embeddings for deleted files in Google Drive.

Credential Setup

• Supabase: Host, service key, and project URL
• OpenAI: API key
• Cohere: API key for reranking
• Google Drive: OAuth or service account setup for API access

Table Structure

• documents table: Stores each chunk and its vector embedding.
• documents_metadata table: Tracks original file, upload time, and document attributes.

Example Use Case

Scenario:
A team uploads project reports to Google Drive. A team member asks, “What are the main deliverables in the Q3 report?”
The agent retrieves, reranks, and cites the relevant Q3 doc’s section, responding instantly in Slack or chat.