📂 Agentic RAG on Google Drive using n8n

An Agentic Retrieval-Augmented Generation (RAG) system built using n8n + Google Drive + Gemini + Postgres, which automatically ingests documents from Google Drive and enables intelligent querying across PDFs, CSVs, Excel files, and text documents.

This system continuously watches a Google Drive folder and keeps the knowledge base in sync whenever files are added or updated.

---

🚀 What This Project Does

• 🔍 Watches a Google Drive folder for file creation & updates
• 📄 Supports multiple document types
  • PDFs
  • CSV files
  • Excel sheets
  • Plain text files
• 🧠 Automatically:
  • Extracts content
  • Generates embeddings
  • Stores structured & unstructured data
• 🤖 Uses an AI Agent to intelligently decide:
  • When to use RAG
  • When to run SQL queries on tabular data
• 💬 Allows querying documents via:
  • Webhook API
  • n8n Chat interface

---

🧠 Why Agentic RAG?

Unlike basic RAG pipelines, this project uses an AI Agent that:

• Understands the type of question
• Chooses between:
  • Semantic search (RAG)
  • SQL queries for numeric / analytical questions
• Avoids hallucination by explicitly stating when data is unavailable

---

🏗️ Architecture Overview

📥 Ingestion Pipeline

1. Google Drive Trigger (File Created / Updated)
2. File download
3. File-type detection
4. Content extraction:
   • PDF → text extraction
   • CSV / Excel → row-wise ingestion
   • Text → direct parsing
5. Metadata storage in Postgres
6. Vector embeddings generation (Gemini)
7. Storage in in-memory vector store

🗄️ Storage

• doc_metadata
  Stores file info, type, and schema (for tabular data)

• doc_rows
  Stores structured rows from CSV / Excel as JSONB

• Vector Store
  Used for semantic retrieval

---

💬 Query Flow

1. User sends a query (chat or webhook)
2. AI Agent:
   • Fetches available documents
   • Decides RAG vs SQL
3. Executes:
   • Semantic retrieval for text-based questions
   • SQL queries for aggregations (avg, sum, max, group by, etc.)
4. Returns grounded, explainable answers

---

🛠️ Tech Stack

• n8n – Workflow orchestration
• Google Drive API – Document source
• Google Gemini – LLM + Embeddings
• PostgreSQL – Metadata & structured data storage
• LangChain nodes (n8n) – Agent, Memory, Tools
• Webhook API – External access

---

🔥 Key Highlights

• Fully automated document sync
• Multi-format document support
• Hybrid RAG + SQL reasoning
• Agent-based decision making
• Production-style data modeling
• Zero manual re-indexing

---

📌 Use Cases

• Internal knowledge assistants
• Company document Q&A bots
• Analytics over spreadsheets using natural language
• AI copilots for operations & reports
• No-code / low-code GenAI workflows

---

📎 Future Improvements

• Persistent vector database (Pinecone / PGVector)
• Access control per document
• Chunk-level citation references
• UI frontend for chat

---

🙌 Credits

Built with ❤️ using n8n and Google Gemini.