📥 Transform Google Drive Documents into Vector Embeddings

Description

Automatically convert documents from Google Drive into vector embeddings using OpenAI, LangChain, and PGVector — fully automated through n8n.

⚙️ What It Does

This workflow monitors a Google Drive folder for new files, supports multiple file types (PDF, TXT, JSON), and processes them into vector embeddings using OpenAI’s text-embedding-3-small model. These embeddings are stored in a Postgres database using the PGVector extension, making them query-ready for semantic search or RAG-based AI agents.

After successful processing, files are moved to a separate “vectorized” folder to avoid duplication.

💡 Use Cases

Powering Retrieval-Augmented Generation (RAG) AI agents
Semantic search across private documents
AI assistant knowledge ingestion
Automated document pipelines for indexing or classification

🧠 Workflow Highlights

Trigger Options:** Manual or Scheduled (3 AM daily by default)
Supported File Types:** PDF, TXT, JSON
Embedding Stack:** LangChain Text Splitter, OpenAI Embeddings, PGVector
Deduplication:** Files are moved after processing
License:** CC BY-SA 4.0
Author:** AlexK1919

🛠 What You’ll Need

Google Drive OAuth2** credentials (connected to Search Folder, Download File, and Move File nodes)
OpenAI API Key** (used in the Embeddings OpenAI node)
Postgres + PGVector** database (connected in the Postgres PGVector Store node)

🔧 Step-by-Step Setup Instructions

Create Google OAuth2 credentials in n8n and connect them to all Google Drive nodes.
Set your source folder ID in the Search Folder node — this is where incoming files are placed.
Set your processed folder ID in the Move File node — files will be moved here after vectorization.
Ensure you have a PGVector-enabled Postgres instance and input the table name and collection in the Postgres PGVector Store node.
Add your OpenAI credentials to the Embeddings OpenAI node and select text-embedding-3-small.
Optional: Activate the Schedule Trigger node to run daily or configure your own schedule.
Run manually by triggering When clicking ‘Test workflow’ for on-demand ingestion.

🧩 Customization Tips

Want to support more file types or enhance the pipeline?

Add new extractors**: Use Extract from File with other formats like DOCX, Markdown, or HTML.
Refine logic by file type**: The Switch node routes files to the correct extraction method based on MIME type (application/pdf, text/plain, application/json).
Pre-process with OCR**: Add an OCR step before extraction to handle scanned PDFs or images.
Add filters**: Enhance the Search Folder or Switch node logic to skip specific files or folders.

📄 License

This workflow is available under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) license. You are free to use, adapt, and share this workflow for non-commercial purposes under the terms of this license.

Full license details: https://creativecommons.org/licenses/by-nc-sa/4.0/